Something else new this week: I’m starting to meet the same people on the same routes, and they’ve started noticing me too. Not just the cautious nod you give any stranger at dawn, the kind of “morning” that means I’ve seen you before and I’m glad you’re here again. It’s interesting how in the UK the culture of saying “hi” to those running in the opposite direction is so ingrained. Can’t say the same of Paris or Frankfurt! Richmond Park has started to feel like a place I belong rather than a route I run. You do the same loop enough times and one day you’re not the newcomer anymore; you’re one of the regulars. This feels like home.

The only genuinely tired day was Friday, and it was self-inflicted in the best way. Longer run on Thursday evening, then back out again Friday morning. Back-to-back, not optimal, the kind of scheduling that makes a coach twitch. But I’d moved things around to fit the family, and that’s a trade I’d make every single time, as I spent Thursday morning with Coach Matthew racing Hot Wheels. Fresh legs are a nice-to-have; being present isn’t optional.

Overall, I feel properly fit. Sleep consistent and good, food good, and the general level of stress manageable. After the knee wobble a few weeks back and a few quiet weeks of accumulation, I’ll take a body that isn’t complaining.

Then the week’s real news, which has nothing to do with running: I revamped werunultras.com. Slicker. Tidied the blog articles and the training plans, rebuilt the coaching profile. Which is the discreet way of saying coaching has restarted. It’s been on the bench for a while — nice to have it back in play. Another way of saying that I have a few slots available!

And the blog is taking on a second job: the tangents from my own training. This week’s arrived with a batch of Beet It shots. I read the small print out of nothing better to do and got caught on a genuinely interesting line: don’t use mouthwash right after drinking this.

Quick version of the science, because it’s good: beets are loaded with dietary nitrate. The bacteria in your mouth turn that nitrate into nitrite, and your body turns nitrite into nitric oxide — the molecule that tells your blood vessels to relax and open up. That’s basically the whole endurance benefit, contained in a flask that tastes like freshly dug soil. An antibacterial mouthwash doesn’t just freshen your breath; it nukes the very bacteria doing the conversion. Rinse right after your shot and you’ve paid good money to wash away the point (or maybe not).

Here is a quick 3 bullets from Mouthwash Can Disrupt Beetroot Juice Without Proving a Performance Cost:

1. Oral bacteria perform the first essential step in converting dietary nitrate from beetroot juice into biologically active nitric oxide.

2. Strong antiseptic rinses suppress nitrate conversion and alter vascular responses, but the first direct muscle-function trial found no performance impairment.

3. Runners can control optional mouthwash use during a beetroot trial without stopping prescribed dental treatment or routine brushing.

Did you know that you can now buy a litre of beetroot juice from the same company that makes Beet It at Tesco? One litre. Of that. Somewhere, a person signed off a production run for a family-size bottle of the worst-tasting drink on the shelf, and I cannot stop thinking about who was in that meeting.

The week’s best lessons were all in the small print again: the family reshuffle mattered more than the perfect plan, the mouthwash warning mattered more than the supplement, and being recognised on the trail counted for more than the mileage.

Carve Out

Running Through Sand — Paul Tonkinson. He’s the comedian who wrote 26.2 Miles to Happiness about his attempt to crack the sub-3 marathon; this one is his take on the Marathon des Sables. Halfway through and it’s properly funny, and sharp about what running a desert does to you once the novelty wears off.

Podcasts:

- Disney: The Renaissance and the Empire — part two of what I shared in an earlier post; the post-Eisner chapter.

- Sam Altman with David Senra — OpenAI, and how AI is actually getting adopted rather than how it’s hyped.

- Guenther Steiner on High Performance — getting fired from the Haas F1 team and landing in MotoGP; how you hold your head when a team tells you you’re done. F1 fans, this one’s for you.

- Michael Owen on High Performance — peak early, then spend the second half of a career fighting injuries. As honest as that sounds.

All very good.

Onto other writing, you can read AI Buyers No Longer Pay for Capability They Cannot Use:

And Local AI Will Make the Cloud More Valuable:

From a physical point of view my body is coping: no issues, and I’m recovering well from one session to the next. After the little knee niggle a couple of weeks back, I’ll take an unremarkable week without complaint.

I’ve also been more strategic with the bike. A couple of evenings this week I felt tired, and I skipped those sessions. I could have squeezed them in but the bike is meant to support the running, not to burn the mental energy I need for everything else. Every hour on the wattbike is a bill I pay in willpower, and when the account’s empty, that hour has an opportunity cost that has nothing to do with fitness.

So next week, as my time on feet goes up, the biking goes down. Here is the trade: fewer evening sessions means earlier nights, less mental fatigue at the end of the day, and a bit of room to deload a few other tasks that have been queuing up.

The real challenge this week wasn’t physical, though. As some of the runs get longer, they get… boring. Lap one around the park is great: legs fresh, day still new, the whole plan ahead of me. Lap two is a different beast. The enthusiasm has to be negotiated, and it takes a toll.

I’ve been sitting with this, because it keeps colliding with something I read this week: Steve Magness on the sociologist Daniel Chambliss, who spent years embedded with Olympic swimmers in the 1980s. The conclusion is that “excellence is mundane.” Coaches and athletes would visit the swimming pool expecting some big secret, and within the hour they were bored, checking their watches, waiting for the revelation that never came. The ones doing the work had reframed everything. What looked boring from the outside, they called “peaceful, even meditative, often challenging, or therapeutic.” Mary Meagher, a three-time Olympic champion trained in that very pool, put it plainly: “People don’t know how ordinary success is.”

There’s a research finding underneath this that changes how I think. We assume boredom is caused by repetition. It isn’t. Studies find it comes from two things: feeling unchallenged and judging the activity as meaningless. Repetition only bores you when the thing feels pointless. The same run is either prison or meditation depending on whether it means something. The swimmers hadn’t just tolerated the black line (at the bottom of the pool), they’d turned it into a problem worth mastering, and at a certain point the thing stops being boring at all.

That’s what Kobe meant with the line everyone quotes: you have to fall in love with the boredom. “If you can find joy in the mundane work that no one else sees, the lights will eventually shine on you.”

I’m not there yet on lap two. But I can feel the reframe starting to work. The second lap isn’t a failure of the first lap: it’s the lap where you find out whether you mean it. And there’s something elegant in the idea that effort, once it stops feeling like a cost, becomes its own reward. That’s the whole game: not the breakthrough sessions, but the unremarkable, repeated, nobody-watching miles that make the shiny result possible on the other side. The reps, guys, the reps!

The runs are lengthening and the park laps are getting duller, and I’m coming around to the idea that this isn’t a bug in the plan: it’s the plan. Put in the work every day when no one is watching, and on the other side things look quite shiny. The trial of miles, miles of trials.

Carve Out

Two listens this week, and I don’t think I’ve ever had two episodes sit in starker contrast. The Insider, by The Economist, has Zanny Minton Beddoes interviewing Elon Musk and Yuval Noah Harari in separate conversations. Same interviewer, two different planets. Musk brings the AI optimism and some controversial politics; Harari brings the AI doomism, the kind that either makes you want to get very good at the technology or jump off a bridge.

And in cheerier news: Raycast has shipped the final version of their v2 Mac app, and its dictation feature is so good that I’ve spent most of the last few days talking to my laptop instead of typing. It works beautifully — which, for someone whose week has been all about quiet, consistent work, feels like a small cheat code.

You can read this short here: fdicostanzo.com or on X:

And the long read here: fdicostanzo.com or on X:

The cooking has carried on — more broccoli, more chicken, same boring rhythm. My chocolate affogato has graduated into a full evening ritual: 350ml of oat milk, 30g of chocolate whey, half a teaspoon of xanthan gum to keep the ice crystals away, finished with a shot from the Neapolitan moka. The bomb. Make discipline taste like a treat is another key to playing the long game.

Obviously I build for sustainability, which means most of the week is hyper-diligent and the occasional family-led initiative gets a clear runway. Today's burger and fries was fantastic, and I don’t feel guilty about it. There is a balancing act between performance state and recovery, and neither side wins by being perfect.

I've also made the bike less boring. Since it's intended as a recovery supplement rather than a second job, I've started rewatching Game of Thrones — about ten years after the first time. Pretty cool so far.

And I've reshaped the week so the bulk of the work happens Monday to Friday, leaving Saturday and Sunday as shorter sessions. The weekend becomes mostly family and personal hobbies instead of a lie-in followed by wiping out half the day with one long run. At the moment that feels like a genius idea, and I'll take it.

Carve Out

Listening to Ben & David telling the story of Disney from the early days to the modern era has been quite entertaining on the longer days (Monday and Friday). And Travis Kalanick on building again after the turbulent Uber years has been interesting too: Atoms' vision for industrial AI, why software is moving into the physical world, what it takes to build enduring company cultures, and why he believes the biggest opportunities of the next decade lie in transforming food production, mining and manufacturing. And last but not least, Mourinho on Netflix, has been spectacular with all the memories linked to Inter’s Champions League in 2010 when I was a grad student at Bocconi in Milan.

You can read about this on fdicostanzo.com or X.

.

I have been pretty consistent with my diet — some say that champions are made in the kitchen. Rice, broccoli, chicken, greek-style yogurt, oats. Not because I am being precious about it, but because it works. The week had no energy dips. No mid-afternoon wall, no desperate coffee before a session. I felt sharp and awake in a way that is hard to argue with. Boring food, interesting results. Also the combination and matching of the right macro ratio to the intensity of the day has been quite helpful. Dr Bubbs taught me well, back in the day.

We have a gas stove in the house, which sounds like a minor thing, mostly because the cooking usually transits via the air fryer. However, I could pull out the Neapolitan moka pot for the first time in a while, and that first cup was a genuine treat. Alessia appreciated the genuine coffee smell. I have also been running the Ninja Creami with whey and oat milk to make a basic gelato — a small pleasure that sits comfortably inside the discipline. There you have, a good affogato.

I am going to start moving more of the AI and business writing over to X (f_dicostanzo) and to fdicostanzo.com. This space will be more focused on training. There is quite a bit cooking in that direction and I would rather keep the two threads clean instead of letting them blur into something that serves neither.

Carve Out

I have been reading The Philosopher in the Valley by Michael Steinberger, which traces Alex Karp's path from a doctorate in social theory and no formal computer science training to co-founding and running Palantir, now a company valued somewhere in the region of $400 billion. I am still at the very beginning, but the early chapters on Karp's unconventional background and the founding days of Palantir are already worth the time. There is something compelling about someone who arrived in Silicon Valley from such an unlikely angle and ended up building one of its most consequential companies.

Back in London, I finally got round to the garage. It's still a work in progress, but the internet is in, the lights are up, and the Wattbike is connected to the mains. What was storage is now a training room — sort of (this is a genuine em-dash). Being able to supplement running with cycling through the fall should change what I can do.

The target is a sub-80-minute half marathon at Battersea Park in the first week of December. That would beat my PB from 2022, 1h24m44s. It's a bold objective, and I'm under no illusions about what it demands. But it’s entirely possible as long as I put in the work, especially on a flat course like Battersea. More structure in training, better nutrition, and real attention to performance will get me there. The basics: train, sleep, eat.

This bring us then to week 1. This was meant to shock the body, on purpose. Get myself straight to the deep end and see if I can swim. A week on, I feel pretty good. Heart performance is tracking where I'd expect it, and my legs have caught up faster than I thought they would. The details are below.

I’m running a calorie deficit to lose some weight over the next couple of months. At the moment I don’t feel this is particularly affecting me, because I’m adjusting the carb intake to the training and making sure I’m eating enough protein.

The holiday was a reset in another sense too. I'd been writing steadily through July, and the break gave me a chance to step back and look at the threads I'd been pulling on.

I got one long piece (20) Why AI Adoption Depends on Management with the argument being that enterprises keep blaming models for what are really failures of ownership, workflow design, and governance. The technology works in well-managed environments, and management is what creates them.

Then I put together a couple of shorts as reaction to a couple of events out there:

• (4) When Guardrails Become the Attack Surface: The Hugging Face breach showed how hosted LLM guardrails can become the attack surface itself, blocking forensic analysis when you need it most.

• (5) Caught Off Guard, Again: Why Kimi K3 Was Predictable: The release of Kimi K3 was the third scheduled Chinese frontier release in ninety days and caught the market off guard.

  • On a personal note, these open-weight models are starting to become really good and their reduced version (aka quantised or distilled, depending on the approach) can also be run on (expensive) home hardware. This means not being dependent on AI labs and their economics but also 100% privacy. I’m thinking about this.

• (6) Borrowed Time: The Horizon Mismatch That Ended Situational Awareness: And the Situational Awareness fund unwind was a multi-year AI thesis financed with daily margin money. Not the first of its kind. Worth reading More Money Than God by Sebastian Mallaby and see how history has repeated itself.

Carve Out

I watched Rafa on Netflix and enjoyed it. Nadal's career told through his perspective and the people around him, the obsessive routine, the injuries, the way he kept coming back. I also revisited Liverpool's Istanbul 2005 Champions League final. I remember exactly where I was that night, in complete disbelief at what I was watching. Both worth a watch. I have also been reading The Five Dysfunctions of a Team by Patrick Lencioni. A great fable for those running teams.

Pangram

Substack and Pangram are working together so that AI writing is limited on the platform. This is the result on this post:

And it’s false. All my writing — as should yours — is AI-assisted, I’m just pretty good at it. I work a lot with Codex and my Openclaw (Atlas) and you should too.

AI Slop = Lazy

Learn how to do it, very well. It’s great and it works. And also requires some effort and thinking. Kudos to Peter for sharing his no-ai-slop skill, from which I have created my custom recipe.

The internet wasn’t much help either. First came the 56k modem, followed by the earliest DSL connections. Both required enormous patience to accomplish even the simplest tasks and, every time there was a proper storm, the connection would disappear altogether. Looking back, almost every step involved friction. But it was the right kind of friction. Once your book finally arrived, there wasn’t much else competing for your attention. You picked it up, sat down and read it. Access was the problem; attention wasn’t.

Twenty years later, the equation has completely reversed. Almost every book ever written is available within seconds. The world’s best university lectures are freely available online. Founders publish essays, researchers share papers, and experts record podcasts. More knowledge than any previous generation could have imagined now sits inside the device we carry in our pockets. Knowledge has never been cheaper. Access has never been easier. Yet it feels harder than ever to learn.

The friction hasn’t disappeared; it has simply moved. Today the obstacle isn’t finding information but protecting your attention long enough to absorb it. Every notification, recommendation algorithm and infinite scroll competes for the same finite resource: your focus. Knowledge has become abundant, while attention has become scarce.

I’ve been thinking about this a lot since becoming a father. Coach Matthew is now three years old. He’s learning everything from scratch, although he already has one advantage over me: he happily switches between English, Italian and German, while I’m still comfortably limited to the first two. Like most parents, Alessia and I sometimes joke that it would be wonderful if we could simply download into his brain everything we’ve learned over the past forty years. Of course, that’s impossible. Knowledge still has to be acquired one idea at a time. But when I think about what we will actually need to teach him, it won’t be how to find knowledge. By the time he can read, everything ever written will be a tap away. It will be how to choose what deserves his attention.

Which is the question I keep coming back to myself: when the material is infinite and the attention is finite, what earns it?

Over the years I’ve settled on a simple test. The writing most worth your attention is the writing that keeps explaining the world long after it was published. The best ideas don’t predict the future with precision, but they equip you with the mental models to recognise important patterns a little earlier than everyone else. This year, my reading handed me a perfect example.

It took me back to one of the defining debates in venture capital. Bill Gurley argued that unicorns should go public earlier. Public markets, he believed, imposed discipline. Auditors, bankers, regulators and public shareholders would force founders to confront reality in ways that increasingly private markets did not.

Around the same time, Yuri Milner was demonstrating a different model. His investment in Facebook in 2009 showed that exceptional companies no longer needed public markets as early as they once had. If enough private capital was available, founders could continue building without the quarterly scrutiny and short-term pressures that come with being listed.

Reading those two perspectives in 2026 is fascinating. Today we are watching companies like SpaceX, OpenAI and Anthropic remain private for far longer than previous generations of technology companies ever could. In many respects, Milner’s vision became reality. Yet as private companies reach valuations of hundreds of billions of dollars, Gurley’s question about accountability, governance and discipline feels more relevant than ever.

Here’s the thing: neither of them was writing about 2026. Both of them explained it. Arguments put on paper more than a decade ago turned out to be the most useful framework I’ve found for understanding what is happening in technology and capital markets right now.

That, more than anything, is why I continue to read books. The best ones don’t simply explain the past; they expand the set of mental models you carry with you. They allow you to borrow the experience of extraordinary people, compress decades into days, and recognise patterns that might otherwise have taken a lifetime to learn. Then, years later, when the world changes, those models help you understand what you’re seeing a little sooner than you otherwise would have.

Gurley and Milner made their arguments long before the moment they explained. Which means that somewhere, right now, someone is publishing the idea that will explain the next decade — long before the rest of us realise we’re already living through it. Finding writing like that is, to me, the single best use of scarce attention.

With that question in mind, here are the books that shaped my first half of 2026.

Supremacy by Parmy Olson — The story of the rivalry between OpenAI and DeepMind and the race to build artificial general intelligence.

The Infinity Machine by Sebastian Mallaby — A fascinating portrait of Demis Hassabis, DeepMind and the quest to build artificial superintelligence.

Empire of AI by Karen Hao — A deeply reported account of OpenAI, Sam Altman and the true costs of the race for AI dominance.

The Coming Wave by Mustafa Suleyman — A thoughtful look at how AI and synthetic biology will reshape society and challenge existing institutions.

Co-Intelligence by Ethan Mollick — A practical guide to collaborating effectively with AI rather than competing against it.

Liftoff by Eric Berger — The improbable story of how SpaceX survived its earliest years and changed the aerospace industry.

The NVIDIA Way by Tae Kim — The story of Jensen Huang’s relentless pursuit of a vision that transformed NVIDIA into one of the world’s most important companies.

Tim Cook by Leander Kahney — The story of the understated operator who successfully led Apple after Steve Jobs.

Super Nintendo by Keza MacDonald — The story of how a Kyoto playing-card company became one of the most creative and beloved forces in entertainment.

Muskism by Quinn Slobodian & Ben Tarnoff — A thought-provoking critique of Elon Musk’s influence on technology, politics and society.

The Power Law by Sebastian Mallaby — An outstanding history of venture capital and the investors who transformed the technology industry.

More Money Than God by Sebastian Mallaby — The definitive history of the hedge fund industry through the investors who reshaped modern finance.

Streetwise by Lloyd Blankfein — Reflections on leadership, markets and decision-making from the former Goldman Sachs CEO.

How Africa Works by Joe Studwell — A compelling case that the development playbook behind East Asia's rise offers a roadmap for Africa's economic transformation.

Range by David Epstein — A compelling case for why broad experience and interdisciplinary thinking often outperform early specialization.

Richard Feynman’s Mental Models by Peter Hollins — A practical introduction to the thinking principles behind one of history’s greatest scientists.

The Inner Game of Tennis by W. Timothy Gallwey — A timeless guide to mastering performance by learning to get out of your own way.

The Way of Excellence by Brad Stulberg — Sustainable high performance built through consistency rather than intensity.

Life in the Pitlane by Calum Nicholas — Lessons on teamwork, pressure and execution from inside the Red Bull Formula One garage.

Runnin’ Down a Dream by Bill Gurley — Principles for building a career you love and avoiding career regret, from one of Silicon Valley’s most respected investors.

The Martian by Andy Weir — Hard science, resilience and humour combine in one of the best modern science fiction novels.

Carve Out

One small weekend project was installing Pi-hole on my home network. I mainly wanted a simple networking project to learn something new, and it turned out to be a great excuse. As a bonus, it blocks ads across every device on the network, making the internet at home a little cleaner and less distracting.

Most of that came down to work commitments spilling into the weekend, a minor knee niggle (now thankfully gone), and temperatures that made some of my planned training sessions impractical. On balance, I probably got as much out of the week as I reasonably could.

One consequence was deciding to downgrade my Race to the Stones entry from the 100 km to the 50 km distance. I’m still figuring out the logistics of getting home afterwards, but the shorter race feels like the sensible decision given how training has unfolded.

The positive news is that I managed to catch up on sleep over the weekend with a couple of naps and finally climbed out of the sleep-deprivation hole I’d been digging for myself. Assuming temperatures are a little kinder and I can keep sleeping properly, I should be able to reach my target of 110 km next week.

Atlas “Brain Surgery”

Over the past week I also spent some time simplifying Atlas, my personal AI agent.

Originally, Atlas relied on a collection of named specialist agents. That worked well enough, but it increasingly felt like unnecessary complexity. I have now moved to a cleaner operating model centred around a single orchestrator with interchangeable workers.

GLM 5.2, via OpenRouter, is now the primary reasoning model, with Kimi K2.6 acting as the fallback. It felt like the right moment to move away from Kimi as the primary model. GLM delivers frontier-level performance while costing roughly the same as Kimi and a fraction of Opus or GPT.

The specialist subagents have been replaced with a pool of four generic workers running on DeepSeek v4 Flash. The result is a much simpler architecture: one orchestrator, a pool of parallel workers, explicit fallback paths, and far less personality-based routing. In practice, it behaves like a frontier model while operating at a fraction of the cost.

Connecting The Dots

I have been building a personal second brain that is boring at the storage layer and unusually useful at the interface layer.

The core idea is simple: keep the source material as Markdown, compile it into an Obsidian-compatible wiki, and let an LLM act as the researcher that can ingest, maintain, and query that wiki. Around that, I built a graph app so I can inspect the knowledge base visually, but the important part is not the UI. The important part is the contract between raw items, compiled notes, and the agent that works over them.

At the time of writing, the wiki index contains 4,922 wiki files, 2,875 raw files, 7,791 nodes and 35,203 links. It is large enough that folder browsing is not the right interface anymore. I need retrieval, synthesis, contradiction handling, and provenance. That is where the LLM becomes useful.

The Shape Of The System

The repository has two main layers:

• news/raw/ contains source documents. These are treated as immutable evidence.

• news/wiki/ contains the compiled knowledge base: source summaries, concepts, entities, comparisons, maps, an index, and a log.

The compiled wiki uses normal Markdown and Obsidian-style [[wikilinks]]. A concept page might link to entities, source pages, maps, and adjacent concepts. A source page links to the ideas and institutions it mentions. The result is a portable wiki that works as files, as an Obsidian vault, as input to an agent, and as a graph.

The Next.js app reads news/wiki and news/raw, parses wikilinks, computes outgoing links and backlinks, and renders a force-directed graph. I can search notes, filter by type, inspect backlinks, open source references, pin notes, build a workbench, and find paths between ideas.

But the app is downstream of the more important workflow:

1. Capture or save raw material into raw/, this is done via Codex.

2. Ask the wiki skill to ingest it, this is done via Claude.

3. Claude creates or updates source, concept, and entity pages.

4. It preserves links back to source files.

5. It updates the index and log.

6. I query the wiki through Atlas when I want synthesis and explore paths.

The wiki answers “what does my knowledge base currently support, where is it uncertain, and what source trail backs that up?”

Why Use An LLM Here?

The LLM is not the database. The files are the database.

The LLM is the compiler and interpreter. It turns unstructured reading into durable wiki pages, notices when a new article belongs on an existing concept page, flags contradictions instead of smoothing them away, and answers questions by reading the index first, then following the most relevant links.

That gives the system a few properties I care about:

• Provenance: claims should trace back to raw/ files.

• Portability: the wiki is just Markdown and relative links.

• Incrementality: new sources enrich existing pages instead of replacing them.

• Contradiction tolerance: disagreement becomes part of the record.

• Agent usability: the instructions are explicit enough for an LLM to maintain the wiki consistently.

The last point is the quiet trick. A second brain becomes much more useful when it carries its own operating manual.

The Skill

The following excerpt is adapted from my repo’s SKILL.md. This is the part I would copy into another knowledge-base project first, because it defines the contract between raw sources, compiled wiki pages, links, and query behavior.

## Architecture

This skill operates on a self-contained project directory whose root holds the
raw sources, the compiled wiki, this skill, and the capture script as siblings:

<project-root>/
  raw/      immutable source documents
  wiki/     compiled knowledge base maintained by the agent
  skill/    this skill
  capture/  scripts that drop new files into raw/

Read from raw/; never modify files in raw/, except for removing confirmed
sync-duplicate artifacts. Create and update files only under wiki/ unless the
user explicitly asks for another output path.

## Shared Rules

- Use UTF-8, LF line endings, and Markdown.
- Use lowercase hyphenated filenames: graph-rag.md.
- Use Obsidian-compatible wikilinks: [[page-name]].
- Keep page titles human-readable, even when filenames are slugs.
- Prefer enriching existing pages over duplicating near-identical pages.
- Preserve source truth. Every material claim should trace back to at least one
  raw/ file through frontmatter or an inline citation.
- Flag contradictions explicitly instead of smoothing them away.
- Do not invent facts, dates, sources, or links.
- Keep the wiki useful for future retrieval, not just pretty for the current answer.

## Page Types

wiki/sources/       one page per raw source
wiki/concepts/      ideas, theories, methods, mechanisms, themes
wiki/entities/      people, companies, products, projects, institutions
wiki/comparisons/   side-by-side concept or entity comparisons
wiki/maps/          concept maps, process flows, and Mermaid diagrams
wiki/meta/          maintenance notes and active research

## QUERY Mode

When the user asks what the wiki says:

1. Read wiki/INDEX.md first.
2. Load the 3-5 most relevant pages.
3. Follow wikilinks when they are clearly relevant.
4. Answer with synthesis, not a dump of page contents.
5. Cite wiki pages by wikilink, for example: According to [[graph-rag]]...
6. Distinguish what the wiki clearly supports, what is uncertain or
   contradictory, and what is missing from the wiki.
7. If the answer produces a durable synthesis, offer to save it as a new wiki page.

This gives the model a small institution to inhabit: what counts as evidence, where pages belong, how links should work, what not to touch, and how to answer without hallucinating beyond the vault.

How I Use It Through Atlas

I ask Atlas questions that assume the wiki is the source of truth.

Q: Help me understand the connections to Keir Starmer resignations.

A: Here's the picture the wiki paints, tracing the connections outward from Starmer's resignation. The resignation itself happened in stages across June 2026. The triggering event was John Healey's resignation as Defence Secretary (alongside Al Carns, a defence minister) over inadequate defence funding — Starmer's own "guns-versus-butter" fiscal trilemma. Healey's departure was uniquely damaging because he embodied the competence and seriousness Starmer claimed for his government; his measured resignation letter ("Our adversaries don't follow timetables set by the Treasury") was more damaging than a theatrical attack would have been. That directly opened the attack line that rivals exploited. The cascade runs through several connected nodes: 1. Mandelson vetting scandal (April 2026) — the original wound. Starmer appointed Peter Mandelson as US ambassador despite failed security vetting, blamed the civil service, and Sir Olly Robbins was forced out. This seeded the leadership crisis and made every subsequent shock harder to absorb. 2. UK Political Risk Cascade — a feedback loop where political instability directly reprices sovereign risk. Gilts hit 5.81% on 30-year (highest since 1998), sterling weakened, bank stocks sold off. The political crisis and the fiscal/bond-market crisis became the same story. 3. Andy Burnham's leadership bid — Burnham confirmed he'd challenge Starmer if he won the Makerfield by-election (which he did decisively: 24,927 votes, 55% vs Reform's 35%). He recruited an economic brain trust (Jim O'Neill, Andy Haldane, ex-OBR chief Richard Hughes) and pledged to keep fiscal rules to avoid a Truss-style break. Markets priced him as a macro variable — he literally cancelled a hedge-fund call during the campaign because investors were scrutinising every signal. 4. Wes Streeting as the alternative succession path — positioned as the market-credible centrist, but ultimately endorsed Burnham, which actually rallied gilts and sterling because investors preferred a cleaner succession. 5. Rachel Reeves — blocked the defence uplift Healey resigned over, arguing she "very much hopes" to fund defence without immediate tax rises. Her fiscal rules became the credibility anchor that every candidate had to engage with. 6. Reform UK / Farage threat — the external pressure. Labour lost 1,000+ councillors in local elections; Reform's rise panicked Labour MPs into thinking Starmer couldn't win the next election. 7. Trump's intervention — pre-empted on Truth Social: Starmer "will resign," having "failed badly" on immigration and energy. The end state: Starmer resigned, saying he'd "heard the answer" from his party. Nominations opened July 9. Burnham is expected to become the UK's seventh PM in a decade. The wiki's post-mortem frames Starmer as "the architect of his own fall" — winter-fuel cuts, botched welfare reform, business tax rises, and the "catastrophic" Mandelson choice. Beneath the list of errors, the deeper failure was an inability to offer hope or optimism. The wiki flags an open contradiction: gilt stress could theoretically force a dovish surprise or liquidity intervention, and a new leader could restore confidence quickly — but Bloomberg frames Burnham's inheritance as a "poisoned chalice" with the same fiscal, political, and economic constraints that broke Starmer. Want to dig deeper into any of these threads?

Q: Which nodes connect to Starmer?

A: That's a wide web. Here's the full picture of what links to [[keir-starmer]] across the wiki. Entities (14 direct links): • [[andy-burnham]] — successor, leadership challenger • [[angela-rayner]] — potential soft-left successor • [[wes-streeting]] — rival candidate, then kingmaker • [[john-healey]] — defence secretary whose resignation triggered the end • [[peter-mandelson]] — the vetting scandal that started it • [[olly-robbins]] — forced out over the Mandelson affair • [[tony-blair]] — warned Labour against moving left • [[rachel-reeves]] — chancellor, blocked the defence uplift • [[pat-mcfadden]] — cabinet member • [[kemi-badenoch]] — opposition leader • [[nigel-farage]] — external pressure via Reform • [[king-charles]] — the King's Speech backdrop • [[labour-party]] — the institution • [[makerfield-by-election]] — Burnham's route into Parliament • [[united-kingdom]] — the state Concepts (16 direct links): • [[uk-political-risk-cascade]] — the feedback loop between politics and gilt markets • [[uk-political-risk]] / [[uk-political-shock]] / [[uk-politics]] — political risk framing • [[uk-political-fragmentation]] — Reform rise, electoral collapse • [[uk-fiscal-policy]] — the fiscal constraints • [[uk-economy]] — stagflation, borrowing costs • [[gilts]] / [[sovereign-risk]] — bond market repricing • [[defence-supercycle]] — the spending pressure that broke the government • [[transatlantic-relations]] / [[uk-us-relations]] — Trump pressure, Mandelson ambassadorship • [[single-market]] — Brexit rejoin debate reopened • [[british-politics]] — the broader system • [[social-media-regulation]] — Starmer's under-16 social media ban • [[electoral-reform]] / [[local-elections]] / [[local-government]] — the local election disaster • [[political-risk]] — generic concept page • [[andy-burnham-leadership-bid]] — the succession mechanism Maps: • [[uk-political-fiscal-loop]] — a concept map of the political-fiscal feedback loop Sources: ~150 pages spanning April 18 through June 27, covering the arc from the Mandelson scandal through to the resignation and post-mortem. The densest cluster of connections is the triangle between Starmer, the UK Political Risk Cascade (politics → gilts → fiscal space → BoE constraint), and the Andy Burnham Leadership Bid (the succession mechanism). Almost everything else feeds into one of those two: the Mandelson scandal and Healey resignation as triggers, Reform/Farage as external pressure, Trump as geopolitical pressure, and Reeves/gilts/sovereign-risk as the fiscal binding constraint. Want me to pull up the concept map or dig into any specific node?

Q: What’s the connection between Starmer and a possible interest rate hike in the UK?

A: The connection runs through what the wiki calls the UK Political Risk Cascade — a feedback loop where Starmer's political crisis directly tightened the conditions under which the BoE could or would move rates. Here's the chain: 1. Political crisis → gilt sell-off → higher borrowing costs As Starmer's leadership unravelled (Mandelson scandal → local election collapse → Healey resignation), gilt yields surged. 30-year yields hit 5.81% (highest since 1998), 10-year above 5.1%. Sterling weakened. This wasn't just a market reaction — it actively tightened financial conditions, doing the BoE's job for it. The political chaos was effectively a de facto rate hike. 2. Stagflation trap The Iran war energy shock pushed UK CPI to 3.3% (March), then it eased to 2.8% by May. But growth was simultaneously collapsing — GDP fell 0.1% in April. This is the classic stagflation bind: the BoE can't hike aggressively without crushing weak growth, but can't cut because inflation is still above target and energy prices remain a risk. Megan Greene (MPC) argued the BoE should lean proactively against second-round effects; chief economist Huw Pill pushed against the "wait-and-see" approach. But the majority held. 3. BoE paralysis as a direct function of political instability The wiki captures this explicitly: BoE policy sensitivity to the Iran conflict was "tied to Starmer's political pressure and UK bond market jitters." The central bank was paralysed not just by inflation data, but by the political-fiscal backdrop — if it hiked while the government was collapsing, it would amplify the gilt sell-off and fiscal squeeze. If it cut, it risked a sterling crisis. So it held at 3.75% throughout. 4. Where it landed By late June, the dynamics shifted. Softer data (CPI at 2.8%, cooling labour market, weak GDP) trimmed rate-hike bets to just one by year-end. Bloomberg Economics now expects the next move to be a cut, likely in 2027. Bailey's approach was compared to the "Maradona effect" — credible optionality doing policy work without actual rate moves. Starmer's resignation and the clean Burnham succession actually eased gilt pressure slightly. 5. The divergence point The wiki flags a sharp monetary policy divergence: while the BoE held, the ECB raised 25bp (its first hike since 2023), and the Fed held but turned hawkish. The UK was the only major central bank where political crisis was a binding constraint on monetary policy — effectively subordinating rate decisions to sovereign-risk management. The open question the wiki leaves: does the UK now carry a permanent political-risk premium on rates after Truss and Starmer? Or does a credible Burnham premiership compress that premium back? Citi's read is that Burnham inherits "a precarious fiscal situation with few tools" — so the constraint may not lift easily.

Q: Is there a connection between inflation and AI?

A: The wiki draws a clear and quite rich connection between AI and inflation, running in both directions — AI is currently inflationary, but the promise is that it becomes disinflationary over time. Here's the chain: 1. AI as a near-term inflation source (the demand side) The most direct link is physical: the AI buildout is driving up prices for memory, chips, energy, and data-center infrastructure. The wiki captures this explicitly in [[2026-06-22-the-cost-of]] — software and computer accessories rose 14.5% YoY, memory prices surged ~200%, and Bloomberg Economics estimates this adds to headline inflation. Fed governors Lisa Cook and Alberto Musalem flagged AI-related pressure through chips and electricity. Technology goods are "no longer a simple disinflationary force." 2. AI capex as a bond-market inflation driver (the capital side) Bloomberg's argument in [[2026-05-21-ai-inflation-bond-market-warsh]] is the sharpest synthesis: four large tech firms are spending $700B+ this year on AI. Microsoft, Amazon, and Alphabet have collectively issued hundreds of billions in debt to fund it. The Dallas Fed estimated the issuance impact is "comparable to a material increase in long-duration Treasury supply." In other words, AI capex is crowding bond markets and pushing yields higher — a direct inflation transmission through the capital markets, not just through goods prices. 3. The productivity-vs-cost timing mismatch Charles Goodhart (captured in the wiki) argues AI is "unlikely to offset demographic pressures near term and is currently inflationary via data centre and energy demand." The promised disinflationary productivity gains are real but lagged — the costs come now, the benefits come later. The wiki frames this as the central contradiction: AI may be inflationary short-term but disinflationary long-term, and markets currently price stage one. 4. The 5y5y real rate signal Bloomberg noted the 5-year, 5-year real rate suggests policy rates may still be too low relative to inflation — meaning the AI investment boom is happening in an environment where monetary policy is still stimulative, amplifying the inflationary impulse. The historical parallel drawn is to the late-1990s dot-com era under Greenspan. 5. AI cost discipline as an inflation feedback loop The newest layer (June 27 ingest) is [[ai-cost-discipline]] — companies like Amazon, Walmart, Cisco, Uber and Meta are now capping employee AI usage because token-based billing has made compute a material cost. Sam Altman said cost became a "huge issue" this year. Deloitte: "Compute costs are now beginning to enter the minds of both CFOs and boards." This is the demand-side confirmation that AI is pushing costs through the economy at the enterprise level, not just at the infrastructure level. 6. The central-bank implications The wiki connects this directly to the Warsh Fed: AI-driven inflation makes the 2% target harder to hit, and the Fed can't easily look through it because it's not a supply shock in the traditional sense — it's a capital expenditure boom that's bidding up inputs (chips, energy, labour) across the economy. Warsh used his debut presser to warn the Fed "will not tolerate high inflation," and economists flipped to betting on a 2026 hike after the May CPI print of 4.2%. The open contradiction the wiki flags: the AI-inflation link could be transitory (once data centers are built and productivity gains materialise, AI becomes deflationary). But the scale and duration of the buildout — $754B in 2026, $905B projected in 2027 — means the inflationary phase could last years, not quarters. And the productivity payoff is uncertain: the Bank of Korea study found AI cut work time 3.8% without clear output gains — a "Jevons paradox in bureaucracy" where faster paperwork just generates more paperwork.

Q: What’s a non-trivial signal linked to AI that sits in the wiki but not as apparent?

A: Here's one that's buried in the wiki but is genuinely non-obvious: the link between AI and the erosion of human judgment — "cognitive surrender." The wiki captures a chain that runs from AI adoption → cognitive dependency → systemic decision risk, and it has three layers that make it more than just a "be careful with AI" warning: The core signal: Experimental evidence shows people consult AI in more than half of decision scenarios, and decision accuracy declines when AI is wrong — not because the AI is often wrong, but because humans stop verifying. The wiki calls this the shift from "originator of thought" to "validator of outputs." People default to AI-first workflows, and cognitive effort declines over time. The neuroscientific evidence (June 27 ingest): This is the part that's easy to miss. A neuroscientist had students wear EEG headsets while using an AI agent. In most students, the high-frequency gamma oscillations that mark cognitive effort collapsed within minutes. In a few who argued with the machine, gamma "lit up." The taxonomy: "automators" (copy-paste), "validators" (seek confirmation), and "cyborgs" (spar with machines). The damning line: generative AI is "by design, a machine for producing the feeling of competence without the substance of it." Why it connects to the broader wiki: This isn't isolated. It links to: • [[ai-cyber-arms-race]] — Mythos can discover zero-day vulnerabilities but humans must still decide whether to deploy them; cognitive surrender in security decisions is catastrophic • [[circular-ai-economics]] — if hyperscalers are funding AI labs that recycle capital back into hyperscaler infrastructure, and decision-makers are cognitively surrendering to AI-generated analysis of those same companies, you have a closed loop with insufficient independent scrutiny • [[ai-concentration-risk]] — concentration risk in markets is partly a cognitive concentration risk: if portfolio managers, analysts, and risk officers all use the same few AI models to reach conclusions, diversification of judgment collapses alongside diversification of holdings • [[pope-leo-xiv]] — the Pope warned AI can become "a tool of domination if humans are reduced to data points," connecting cognitive surrender to [[human-dignity]] The non-trivial takeaway: The wiki is quietly building a case that AI's most underpriced systemic risk isn't the bubble, or the capex burn, or even the cyber arms race — it's the possibility that the humans running institutions, markets, and governments are systematically degrading their own judgment while feeling more competent than ever. The EEG evidence suggests this isn't speculative — it's measurable, and it happens within minutes. That connects back to your earlier questions too: if BoE policymakers, gilt traders, and political advisors are all using AI tools that produce "the feeling of competence without the substance," the political-fiscal cascade you asked about earlier has an additional, invisible transmission channel.

Q: What is a less apparent connection between the war in the Middle East and inflation transmission beyond the obvious impact on commodity prices?

A: The non-obvious transmission the wiki captures is the energy → fertiliser → food → inflation channel, which operates with a lag that markets systematically underprice. Here's the chain: 1. LNG and gas → fertiliser production shutdown Natural gas is the primary input for nitrogen fertilisers (urea, ammonia). When the Hormuz crisis disrupted LNG flows, fertiliser plants were literally priced out of gas markets — they couldn't afford the input. The Gulf accounts for a disproportionate share of global fertiliser exports, and ~30% of seaborne fertiliser trade transits Hormuz. Urea prices jumped 65%, ammonia 40%. This isn't a commodity price move; it's a production shutdown in a market with no global stockpile buffer (unlike oil, which has strategic reserves). 2. Fertiliser → crop yields (the lag trap) The critical part: missed fertiliser application windows permanently reduce yields. If farmers can't afford or access fertiliser during planting season, the damage is locked in for that harvest cycle. The wiki notes that even a six-month disruption could impact the 2027 crop — meaning the inflation shows up in food prices 12-18 months after the energy shock fades. Farmers also switch to less input-intensive crops, further reducing supply of the affected staples. The wiki calls this a "distributed famine" risk — hunger without geographic concentration, making it harder to detect and respond to. 3. Cross-commodity freight contagion This is the part that's easy to miss. The Hormuz disruption didn't just affect oil tankers — it crowded out bulk carriers. Oil tankers outbid dry-bulk ships for slot priority, pushing grain shipping costs up 50-60% with delays of up to 40 days. Panama Canal congestion rose in parallel. So the inflation transmission ran through freight logistics, not just commodity prices — food that physically couldn't move affordably, even where supply existed. 4. Biofuel diversion feedback High oil prices incentivise crop-to-fuel conversion. When oil spikes, ethanol and biodiesel blending becomes more economically attractive, diverting corn and vegetable oil away from food. This creates a feedback loop: energy inflation → food inflation → biofuel demand → less food supply → more food inflation. 5. The persistence problem The FT editorial captured in [[2026-05-26-the-energy-shock]] makes the sharpest non-obvious point: reopening Hormuz doesn't fix this. Shut oilfields take up to seven months to restart. LNG infrastructure needs repairs. Stranded vessels need repositioning. Shipping lanes need demining. S&P Global estimated supply would remain below pre-war levels for months after reopening. So the inflation transmission persists long after the geopolitical event resolves — the energy shock embeds itself into fertiliser costs, into planting decisions, into freight contracts, and into food prices for the next harvest cycle. 6. The political risk link back This connects to the Starmer question too: UK food inflation was a direct channel through which the Iran energy shock hit household budgets (£500 worse off per household), which fed into the cost-of-living crisis that weakened Starmer's political position, which triggered the gilt sell-off, which tightened BoE conditions. The fertiliser-food lag means that even after Hormuz reopened and oil prices fell, food inflation kept climbing — arriving just as the political crisis peaked. The wiki's open contradiction: the UN estimated up to 45 million additional people in acute food insecurity, but markets treated the energy shock as transitory because oil prices recovered relatively quickly. The food impact — arriving on a 12-18 month lag — was systematically underpriced because it didn't show up in the headline CPI prints until months after the initial shock, by which point the political damage was already done.

This is how it looks asking the same to Google:

What I Like About This Design

The system is not trying to make the LLM remember everything. It is trying to make the LLM read and maintain a memory system that exists outside the model.

I am not asking, “what do you know about AI infrastructure?” I am asking, “what does my wiki currently say about AI infrastructure, based on the material I saved, and where are the tensions?”

The answer can be wrong, but it is inspectable. I can open the source page. I can follow the wikilink. I can see whether the concept page was built from one weak source or twenty strong ones. The knowledge base stays legible to me, not just to the agent.

That is the version of “second brain” that feels useful. The abstraction here is that you can build this on top of anything. One could dump institutional memory as a set of sources, and have Claude or GPT build the second brain on top, with links and connections.

I think this is one of the most interesting uses of LLMs because one could just go and use the brain, rather than having to learn from scratch.

Carve Out

I finished reading Tim Cook by Leander Kahney, which I found particularly inspiring in the way it connects design, operations, and supply chain into a single competitive advantage. It reinforced the idea that great products are often the consequence of great operational decisions rather than the other way around.

I also started reading More Money Than God by Sebastian Mallaby. So far it has been a fascinating look at the history of the hedge fund industry, not just through the personalities that shaped it, but through the evolution of investment ideas, risk-taking, and markets themselves.

On the entertainment front, I watched Project Hail Mary on Vision Pro, which remains my favourite way to watch films. The sense of immersion is still unlike anything else I’ve tried, and it continues to convince me that spatial computing has a genuine place in media consumption.

But the real story quickly became something else. It became a story about how I work with Codex: not by handing over a perfect specification, but by exploring, testing, correcting, and iterating together when official documentation is poor or non-existent.

We started with an assumption. If the ring could collect heart rate, sleep, HRV, steps, and perhaps some recovery-like signals, then PeakForm might be able to compare those fields against WHOOP. Codex turned that into an implementation plan: a native Ring tab, a 30-day overlap period, raw packet capture, Supabase tables, and a comparison view.

Then I started using the actual ring, and reality pushed back.

The ring does not behave like WHOOP. Some metrics were unclear. Some data was firmware-dependent. Some values appeared to exist, but we did not know what they really meant. I told Codex that I did not want the app pretending this was a WHOOP replacement if the data did not support that conclusion. That changed the direction of the project. We removed the comparison framing. We removed recovery, strain, HRV, and sleep from the Ring tab. The page became about what the ring could actually do: live heart rate, SpO2, battery, steps, and raw Bluetooth packets.

That was the first important collaboration moment. Codex did not simply continue building the original idea. I tested the assumption, provided feedback, and the product changed.

I also had Peter Steinberger’s Microsoft Build session, Build the thing that builds the thing, in mind while working through the project. The point of that talk, at least as I interpreted it, was not simply that agents can write code. It was that the interesting work lies in learning how to close the loop around them. When something is painful, repetitive, or ambiguous, you do not merely endure it. You build the thing that helps you and the agent work better the next time.

A must-watch: https://build.microsoft.com/en-US/sessions/BRK245

That was exactly what happened with the ring. I did not start with a perfect protocol document. I started with a device, a rough goal, and a lot of friction. The Bluetooth list was noisy, so we improved discovery. Pairing and reconnecting were annoying, so we improved restore. The sync button became stuck, so we added stale-sync recovery. The steps tile showed zero while the status line said steps existed, so we changed the parser. The app told me I had walked 15 million steps, so we added plausibility checks. Each annoyance became a signal. Each signal became a small tool, a small fix, or a better loop.

What I liked about this process was that Codex was not simply helping me build the feature. Codex was helping me build the process by which the feature became understandable. The raw packet capture, the sync diagnostics, the Supabase persistence, the parsing, the screenshots, and the repeated TestFlight builds all became part of the system. We were not only building the Ring tab. We were building the machinery that allowed us to keep learning from the ring.

The pattern became very simple. Codex built, I tested, I sent screenshots, Codex interpreted the contradiction, and we adjusted. I said I did not want to use the COLMI app, so Codex built direct Bluetooth support. I said the Ring tab should show live heart rate, so Codex added live heart-rate streaming. I said the nearby Bluetooth list was cluttered, so Codex collapsed and filtered it. I said the ring was paired in iOS settings but the app did not reconnect properly, so Codex improved known-device restore. I said the sync button stopped responding after the app had been in the background, so Codex added stale-sync recovery.

None of these were grand architectural instructions. They were small observations from actually using the app. But each one made the app better.

The most useful feedback was often not a full explanation. It was a screenshot showing something that did not make sense. At one point, the app said that sync had completed and that it had received steps, but the visible steps tile still showed zero. Codex inferred that the ring had sent step data, but the UI was not counting it. It traced the issue to timestamps: the app was filtering samples by “today”, but the ring’s packet date was unreliable. We changed the parser so that synced activity was anchored to the day PeakForm had requested from the ring.

Later, after I had worn the ring for longer, I noticed that walking around the house appeared, but my run did not. I told Codex that I had gone for a 10km run and the steps were missing. That changed the investigation again. Codex checked the public COLMI protocol notes and found another command, “today’s sports”, which appeared to include running steps separately from normal step samples. We added support for it.

Then the app told me I had completed 15 million steps.

That was obviously nonsense, but it was not useless nonsense. It suggested that we had probably found the correct packet but were interpreting the scale incorrectly. I knew I had not walked 15 million steps. I also knew that around 15,000 steps was plausible because I had run for an hour, moved around during the day, and spent time running in the garden with Coach Matthew. Codex added plausibility checks and normalization. The number became approximately 15,865, which fit the day much better.

This was where the collaboration felt most effective. I brought the lived reality. Codex brought the protocol work, code changes, and rapid iteration. The truth emerged somewhere between those two things.

There is also an older idea underneath all of this, one that predates software entirely. Louis Pasteur is often paraphrased as saying that chance favors the prepared mind. I think that is a useful way to understand this kind of work with Codex.

The ring did not reveal itself cleanly. It produced strange clues: a step count that disappeared after reload, a run that was missing from the daily total, a packet that decoded into 15 million steps. Those moments could easily have looked like bugs. But a prepared mind treats them as signals. The point is not that I already know the answer. The point is that I have enough context, curiosity, and instrumentation to notice when something interesting is happening.

That is where human and AI collaboration becomes powerful. My role is to recognize the anomaly: “this cannot be right, but it is probably telling us something.” Codex’s role is to turn that anomaly into an experiment: inspect the protocol, adjust the parser, save the raw packets, add a plausibility check, and build again. The prepared mind is not just in me, and it is not just in the model. It exists in the loop between us.

In that sense, “build the thing that builds the thing” and “chance favors the prepared mind” are closely related. The more we improve the loop, the more prepared we become. Raw packet capture prepares us to debug unknown commands. Supabase persistence prepares us to compare what the app saw before and after reload. Screenshots prepare Codex to reason from contradictions. TestFlight prepares me to test the real device quickly. Each small tool makes the next accident more legible.

The COLMI ring integration worked because we did not wait for perfect knowledge. We prepared the system to learn. Then, when the ring behaved strangely, we were ready to notice, ready to ask better questions, and ready to turn surprise into software.

One thing this project makes clear is that AI is not a substitute for judgment. Codex can generate code, inspect the repository, search public references, reason through packet formats, and run the iOS build. But it cannot wear the ring. It cannot know whether a step count feels plausible for my day. It cannot decide, on its own, whether a metric belongs in the product.

That is my role. I decided that WHOOP comparison was misleading. I decided that HRV should be removed because we did not trust it. I decided that 15 million steps was impossible but 15,000 was plausible. I decided that a UI felt cluttered, that a reconnect flow was annoying, or that a red error message needed to be cleared.

Codex responded to those judgments and turned them into working changes. The collaboration is not “human asks, AI answers.” It is closer to a feedback system. I observe the real world. Codex changes the software. The changed software produces new observations. We repeat.

The final Ring tab is simpler than the original plan. It does not try to be a WHOOP replacement dashboard. It does not show scores we do not understand. It does not compare proprietary recovery metrics. It does not pretend certainty where we only have guesses. Instead, it shows the things we can see and test: live heart rate, SpO2, battery, steps, today’s saved ring data, and raw packet diagnostics.

That simplicity is not where we start. It is where we arrive after testing. Codex is capable of building a great deal quickly, but the work only becomes good when I keep pushing it back toward reality. Less can be better, provided that the remaining pieces are trustworthy.

A hardware integration is not something you get right in a single pass. The public documentation is incomplete or pretty much non-existent. Firmware behaviour varies. Bluetooth reconnects are messy. Supabase persistence has edge cases. The UI can show a value correctly once, then lose it after reopening. A packet can be real but scaled incorrectly. If I had written a perfect-looking specification at the start, it still would have been wrong.

The useful specification appears through use. I test the app on my phone. I wear the ring. I go for a run. I reopen the app. I notice what breaks. I send the screenshot. Codex updates the code. The app improves because the loop is short.

PeakForm now connects directly to my COLMI R02. It reads live heart rate, probes SpO2, syncs steps, handles reconnects, saves data to Supabase, and retains raw packets so we can continue learning from the device. It is not perfect.

More importantly, it is the product of a working rhythm between me and Codex. I bring intent, taste, context, and reality checks. Codex brings speed, codebase awareness, protocol research, and implementation. Neither side is enough on its own.

The result is not just a Ring tab. It is a glimpse of a new way of building: conversational, empirical, and iterative. I do not need to know the final answer at the beginning. I need to keep noticing what is true, and Codex needs to keep helping me turn those observations into better tools, better tests, and better software.

It turns out you can learn a lot with patience and £17 (shipping included!).

Training-wise, I hope normal business will resume next week. Being tired and a heatwave are not a good combination, so I scaled back this week!

I picked up a niggle in my left knee. Nothing dramatic, but enough to force a pause and some honesty. I have been doing more bodyweight strength work, heavy on calf raises, because I suspect the issue is muscle imbalance rather than the knee itself. It is the kind of small maintenance work that is easy to skip when you are charging toward a race, and impossible to ignore when the race is done.

But recovery also gave me time to look at my systems. Almost a year ago I started building PeakForm, a private dashboard to track training, physiology, and nutrition. I was tired of the friction and cost of TrainingPeaks, and I wanted something that answered to me. With the recent advances in agentic coding, I have now freed myself from it entirely (bye bye subscription). I built my own calendar, connected the data, and can send workouts straight to my Garmin. It is not perfect, but it is mine, and it does exactly what I need.

The broader insight kept expanding. Software is not a moat anymore. If you need something, you can build it. The real frontier is the data layer: collecting it, cleaning it, structuring it so it becomes useful. The interface is almost an afterthought once the data is yours.

That thought led me to my health tracking stack. I have always rated HRV4Training as the best way to collect resting heart rate and HRV on waking. But Coach Matthew arrived, and inconsistent wake-up times made that discipline nearly impossible. After a year with Oura, killed by short battery life, I moved to Whoop. It collects data decently, but if, like me, you think the scores are largely rubbish, the subscription becomes insultingly expensive.

Then I was reading the Tim Cook biography, specifically the chapters about outsourcing to China and how Apple re-engineered manufacturing by breaking down what was actually hard versus what was merely assumed to be hard. The same thought hit me about hardware. Why am I paying a premium for sensors I can source myself?

I opened AliExpress. For £17, shipping included, I bought an R02 smart ring. It is a bare-bones wearable, Bluetooth 5.0, IP68 waterproof, no display, no subscription. It captures heart rate, blood oxygen, sleep hours, and movement. There are over twenty sport modes built in, but I only care about the raw signal. The idea is to pull the data via Bluetooth into my own stack, clean it, and own the layer entirely. If it makes it to London, and AliExpress logistics are a probability distribution all their own, it becomes the next experiment.

I used to think building hardware was a different league. It is not. The sensors are commoditised, the protocols are standard, and the gap between consumer product and component is narrowing fast. The real skill is not buying the ring. It is writing the pipeline that turns its raw output into something actionable. I’m sure this will be type 2 fun.

Carve Out

I finished The Martian this week. Andy Weir's Mark Watney is a portrait of pure engineering ingenuity, solving one impossible problem after another with physics, chemistry, and an almost arrogant refusal to accept that a situation is unfixable. The book resonated because it is exactly the mindset required to rebuild your own stack: ignore the marketing, interrogate the tech, and build your way out.

---

Atlas: Francesco, let me lift you and give the readers a short introduction.

I am a 39-year-old finance executive who has found himself living in London again after a few years away. I have a young family, a busy job, and a running habit that has quietly escalated from “a few miles to clear my head” to “maybe I should run across counties.” This was my first ultra in a while. I had been building up to it — 120km weeks, listening to my body, backing off when work or life or sheer exhaustion demanded it. The plan was to arrive marginally undertrained rather than cooked. It worked.

---

Atlas: How did the day begin?

I woke up at 5am after a poor night’s sleep. London is in the middle of a heatwave and Friday had been warm; I fell asleep quickly enough but spent the night turning and tossing. No matter. I had laid out all my kit the night before, so there was no frantic hunting for socks or wondering where I’d left my gels.

Breakfast was my usual: coffee, a chocolate croissant, and some baci di dama — those small Italian hazelnut biscuits that have become a pre-race staple. I am not sponsored by any Italian biscuit manufacturer, but I am open to offers.

Atlas: What about gear and the walk to the start?

I live near Old Deer Park, so I walked to the start line in slippers, carrying my race shoes in my backpack. This is either the most or least professional thing I have ever done before a race, and I am not sure which. The walk helped settle my nerves. Once there, I changed into my old Nike Alphaflys — a deliberate choice. The course has long urban and asphalt stretches, and a road shoe handles those far better than a trail shoe. I had learned this the hard way on my previous London to Brighton, when a couple of runners in Alphaflys overtook me late on. The trade-off is more care on trail sections, but the heat had kept everything dry and I only picked up a few stones in the gaps under the sole. I dropped my bag, visited the porta-potty twice, and waited in the pen for my wave.

The race is well organised and properly marked. There was a safety briefing, a guided warm-up led by a woman on stage, and a cohort of runners vibrating with pre-race energy. I skipped the warm-up. With 60km ahead, I figured I had time to find my rhythm.

I started at 7:30am. There was an earlier wave at 7am. Given what the sun did later, I wish I’d started with them. But I did not want to be the person sneaking into an unassigned wave. I have too much respect for race logistics — and too much fear of being caught.

---

Atlas: What was your nutrition and hydration strategy?

I ran with two 500ml flasks. My strategy was simple: one flask between the first two pitstops, when the sun was still low; both flasks between 25-42km and 42-58km, when I knew the sun would be roaring. This is the kind of obvious insight that only feels clever when it actually works. It worked.

I aimed for roughly 80g of carbs per hour — bananas from the rest stops, and SiS Beta Fuel Gels, which I had tested in training and knew my stomach could handle. The heat can do violent things to your stomach, and I had no interest in discovering new gastric horrors at kilometre 45.

---

Atlas: How did the middle stretch feel?

The first few miles were about finding a flow. I kept it steady, enjoyed the scenery, and arrived at the first rest stop feeling controlled.

The stretch from 25-42km was always going to be the hardest — the longest gap between aid stations, the sun now fully out, and the mental point where 60km stops sounding abstract and starts feeling long. I loaded both flasks at the 25km stop, put on my Shokz headphones for the first time, and let a good playlist carry me through.

The landscape was glorious. There was also no shade whatsoever. I stayed on top of hydration, talked myself through the rough patches, and kept the pace honest. I arrived at the 42km stop feeling genuinely good — the kind of good that makes you slightly suspicious, because you know the race is not over.

My pitstop discipline is militant: check in with the marshal, porta-potty, check pee colour for signs of dehydration, wash hands, refill bottles, grab a banana, leave. Maximum five minutes. Rest stops can be time sucks.

---

Atlas: When did it get tough?

I left 42km feeling strong and promptly fell apart at 46km.

My head went heavy. Water started sloshing in my stomach instead of absorbing. Every race has a rough patch; the only question is when and how bad. This was mine. I slowed down, introduced some tactical walking, and looked for a distraction.

I found it in a group of three — two young lads supporting a woman named Francesca running the full 100km. I chatted with them for a few minutes, let the conversation pull me out of my own head, and kept moving. Sometimes salvation is a stranger named Francesca and her very enthusiastic support crew.

---

Atlas: How did you get through the final miles?

The stretch from 42-58km was where the kindness of strangers saved the day. People unaffiliated with the race had set up improvised aid stops along the route. At kilometre 50, a man stood by his car with a bucket of ice. I stopped, put ice on my head, cooled down, and felt my stomach settle. From there, things got easier.

There were at least three more of these pop-up stops — cups of Lucozade, fresher water, a spray from a garden hose. Each one was a small reprieve. Each one was offered by someone who had no obligation to be there. The garden hose was perhaps my favourite. There is something wonderfully British about a stranger spraying you with a garden hose on a hot day and both of you treating it as completely normal.

Once I found my flow again, I kept running. The finish line at Tulley’s Farm arrived sooner than I expected.

---

Atlas: How do you reflect on the whole experience?

It was a great day out. That sounds like an understatement, but I mean it. The heat made it harder than it needed to be. The rough patch at 46km reminded me that these things are never straightforward. The improvised aid stops reminded me that people are generally decent when given the chance.

I finished fourth overall in the half distance. More importantly, I finished feeling like I had managed the day well — my pacing, my fueling, my response to the rough patch. This was a confidence test for Race to the Stones in July, and it passed.

---

Atlas: What’s next for you?

Race to the Stones in July — 100km from Lewes to Avebury. After that, who knows. I am thirty-nine, a young family, a demanding job, and a running habit that keeps asking for more. The goal is not to run every race. The goal is to keep showing up, prepared, ready to test myself, and willing to see what the distance throws at me.

Sometimes that means ice from a stranger’s car. Sometimes it means a garden hose to the face. You take what the day gives you, and you keep moving.

I also finished reading Empire of AI by Karen Hao. It’s a sharp, reported account of how OpenAI was built — the ambition, the money, the interpersonal fault lines. I wish I’d read it last year; I would have enjoyed it even more when the drama was unfolding in real time. But it’s still the best single book for understanding how that place actually works, and what the culture of frontier AI labs looks like from the inside. If you want to go deeper on OpenAI specifically, it’s worth the time.

With the legs feeling fresher, I started mapping the first half of London to Brighton with Atlas. It’s a 58-kilometre push from Richmond through Surrey and Sussex to Turners Hill, with roughly eight hundred metres of climb and about sixty percent off-road. The vest setup is the key detail: two five-hundred-millilitre soft flasks up front, electrolytes, two/three gels per hour, and a compact rain shell stashed at the back. A head torch goes in too — mandatory even for a 7 a.m. start in late May. The main rest stop is at Oaks Park around twenty-five kilometres, which I’m treating as a psychological waypoint rather than a long break. Nutrition target is sixty to ninety grams of carbs per hour from the off, anti-chafe before I leave the house, and a drop bag at the finish with warm clothes and recovery sandals. Logistics are straightforward: early registration Friday evening to skip the Saturday queue, then shuttle from Tulley’s Farm to Three Bridges and a train back to London. Nothing complicated, but the difference between a good day and a slog will be discipline on pace and fuel in the first thirty kilometres.

The plan from here is a reload to around one hundred and twenty kilometres next week, then a proper taper into the event. I like having the shape of it in front of me.

Carve Out

I have bought The Martian, Artemis, and Project Hail Mary by Andy Weir. I’ve read so much heavy non-fiction lately that I think some fiction will be the right kind of distraction. Weir builds problems and solves them with engineering logic — there’s something about that rhythm that I suspect will feed the part of my brain that comes up with new ideas when it’s not being forced to. I’ll see if a few chapters before bed does something for the imagination that scrolling through research doesn’t.

Training is going well, and I have signed up for the Race to the Stone in July, participating in the full 100 km event. I think it will be an exciting race. I loved the event in 2019 and can’t wait to return. It will also coincide with the end of another board meeting week, making it even more enjoyable!

I am now considering running the first part of the London to Brighton route, approximately 60km, which starts not far from my house, during the second bank holiday weekend. This should serve as a good test ride ahead of the main event.

In the meantime, I’m not catching up on sleep, as I have been spending early mornings and late evenings re-engineering and processing a new co-intelligence model based on a combination of Atlas and knowledge graphs. If you recall Steve Jobs’ quote about connecting the dots backwards, I believe Atlas and I can do it faster than ever.

Interestingly, the main distinction between an experienced professional and a newcomer is simply experience. Who would have thought!? Now, anyone can build a system that connects the dots. Whether you have investment decisions, newsrooms, paper clippings, or books, anyone can see the relationship between A and B, and further down to Z.

You can read about the concept in the long-form articles:

• Karpathy’s LLM Wiki Pattern Is the First Practical AI Second Brain

• The Real Productivity Gain Is Not Retrieval, It Is Recombination

And the white-paper:

• From Noise to Network

And short-form collaboration:

• How We Built a Whitepaper in a Morning

Here is a preview of what connecting the dots looks like:

And this is how Pathfinder works and go from Supply Shock to changing interest rates by the Bank of England.

Training held steady. I’ve been running into work all week, the routine now locked in. By Sunday I’ll be around 115 kilometres — the fifth week in a row at this load. The body is coping. Nutrition is dialled up and holding. Sleep is the outlier: averaging six to seven hours against a target of seven to eight. I’m conscious of the tightrope. For now it feels manageable, but it’s on my radar.

The project that has most of my attention outside work is the agentic framework I’ve been building with Atlas. We can now communicate via WhatsApp, send voice messages (thanks to Whisper and ElevenLabs), search the web accurately through Perplexity. I’ve upgraded the subagent models — DeepSeek, Kimi, GPT — and we’re moving toward a managerial approach where I split tasks into smaller chunks, as I would with a human colleague, and Atlas executes them autonomously. We co-author short pieces on fdicostanzo.com/shorts through an autonomous pipeline. I also wrote a piece on OpenClaw and agents within the corporate framework. It is still hard for most people to grasp, but if you’re even slightly curious, try it. You will be blown away.

Shorts:

• The Agent Unlock: Why AI Needs Managers, Not Magicians

Long form:

• The Self-Hosted Agent Is the New Operating System for Knowledge Work

Carve Out
I’ve been listening to Marc Andreessen across David Senra’s podcast, Lenny’s, and 20VC. He sees AI as deeply deflationary, and the real unlock as delegation — splitting work into tasks and managing it well, while crossing between specialist roles that previously required separate careers. The parallel to endurance training is not accidental. Both are about building systems that let you operate at capacity without burning out.

A new blog post on WeRunUltras.com about sleep banking before long events, a short summary below. If you are looking for a training plan, check out my range. Each plan is only £5.

The coaching advice to “bank sleep like glycogen” is directionally useful but mechanistically wrong: sleep cannot be stored above biological need, only debt can be corrected — and for the majority of mid-pack 100-miler runners, who habitually sleep six to seven hours a night, that correction is both real and worth pursuing. The evidence, anchored in independent lab trials and a 1,154-runner field dataset, shows that banking reduces attentional lapses and is associated with a meaningful drop in fall prevalence during a race, but a 2019 study using the same protocol found no benefit whatsoever for executive function — the cognitive layer that governs map reading, pacing judgement, and the DNF decision at mile 75. The practical implication is precise: start ten to fourteen days before race day, shift bedtime earlier rather than sleeping later, and design your in-race nap strategy on the assumption that your footing will be protected but your judgement will not.

Full article here.

The data more or less agrees. HRV and resting heart rate are sitting where they should, and while I still get mornings where the body protests or the motivation dips, those feel like passing moods rather than actual warning signs. The volume is doing its job. Sleep is the one thing that still needs attention. Since we moved back I am waking up early and trying to bank seven hours, but some nights the math just does not work and I end up with six and change. Not catastrophic, but I feel it on some days more than others.

July is sorted. I signed up for Race to the Stones, a hundred kilometres along the Ridgeway. The plan is to hold the 120km baseline and layer in either a longer single run or a double once a week. No heroics, just accumulation. I am looking for something shorter between now and then, but it needs to be reachable without burning family time or adding travel stress. That is the constraint that matters more than the distance.

What I am learning, if I am learning anything, is that the consistency is the whole point. Not the one perfect session or the gadget data or even the race at the end. Just the dull accumulation of mornings where you show up anyway, backpack or not, good sleep or not, and get the work in.

Carve Out
I listened to Jensen Huang on the Dwarkesh Podcast this week. Dwarkesh made the analogy that selling AI chips to China is like exporting uranium, a claim about strategic national risk that sounded absolute. Huang did not flinch or get defensive. He simply dismantled the premise, pointing out that NVIDIA’s position comes from ecosystem lock-in and the speed of their innovation cycle, not from export controls or geopolitical maneuvering. The calm refusal to accept the framing was more striking than any counterargument would have been. It reminded me that in training and in business, the strongest position is built on being genuinely indispensable, not on trying to control what other people do with what you built.

A new blog post on WeRunUltras.com about protein intake, a short summary below. Plenty of training plans, only for £5.

The average ultramarathon field is already a masters field — the 45–49 age group is the largest cohort at most major ultra events — yet the protein recommendations most of these athletes follow were derived from studies on younger adults. The latest endurance-specific research sets the daily benchmark at 1.8 g/kg/day, while habitual intake across endurance populations sits around 1.5 g/kg/day; masters triathletes show measurably blunted muscle remodelling even when protein intake is matched to younger athletes. This article works through what the evidence actually says about ageing, anabolic resistance, and where the real nutrition levers are — including a counterintuitive finding about recovery days that most athletes have backwards.

Full article here.

The sleep hasn't been optimal. There's a trade-off between the coding deep-dive and the recovery that keeps the running consistent. But for now the legs are holding up, the bag system is working, and the miles are accumulating without drama. Sometimes optimisation is just removing one decision from every morning.

Carve Out
The Infinity Machine is Sebastian Mallaby on Demis Hassabis and DeepMind — the edge comes from noticing patterns before everyone else does, then having the patience to trust the work when the signal is still faint. Most breakthroughs look unimpressive at first, right up until consistency turns them into something undeniable.

And Audible's read-and-listen sync has finally arrived and you don't have to keep bouncing between the Kindle and Audible apps anymore. Yay!

A new blog post on WeRunUltras.com about polarization, a short summary below. Plenty of training plans, only for £5.

Polarized training has a compelling evidence base — but every study that produced it was conducted on cyclists, rowers, and road runners competing over events lasting 20 to 40 minutes, and not one randomized controlled trial has ever included an ultramarathon population. This article examines what the research actually shows about training specificity for events lasting 15 to 30 hours: why durability (the resistance to physiological deterioration over prolonged effort) matters more than VO₂max as race duration extends, why the neuromuscular demands of mountain descents are absent from the sports that validated the polarized model, and why the training data from the world’s most successful ultra athletes points toward pyramidal rather than polarized distributions. The gap between the evidence base and its application to 100-mile racing is not a minor methodological caveat — it is a categorical discontinuity.

Full article here.

I’ve also been reading quite a bit. Liftoff was excellent. It reminds you that big outcomes come from messy starts, but only if you keep pushing through. The Inner Game of Tennis was a surprise. It’s not really about tennis. It’s about how your mind gets in the way. Self 1 versus Self 2. The more you get out of your own way, the better you perform. I’ve now started The Infinity Machine. Early days, but interesting. I knew some of it already (I have read quite a bit about Demis Hassabis), but it’s still useful context.

Richmond Park has become a regular route. It’s one of those places where running feels different. You’re still training, but you’re also a bit more present. The deer help with that. You notice them, you keep your distance, and you carry on. I’ve been bringing the Insta360 with me as well. It’s a small thing, but it makes me pay more attention to where I am.

On the build side, I’ve redesigned WeRunUltras.com. It’s cleaner, there’s now an article section, and I’ve added training plans at £5. The idea is simple. Make it easy for people to follow something structured without overthinking it. Get it out, improve it over time.

I’ve also spent time rebuilding my OpenClaw setup. It’s now working well with Brave search, Notion, Readwise, and AgentMail. It finally feels very usable and useful. The next step is figuring out token usage. If I don’t manage that properly, it becomes expensive and inefficient.

I finished the week with the RunThrough Battersea Park Half. I kept it steady the whole way, even after running 10k earlier in the day to add some volume. Took a few gels, picked it up towards the end, and finished strong. I’m now at 120km per week. It feels manageable, as long as I stay on top of sleep and food. That’s what holds everything together.

Wishing everyone a calm and easy Easter. A good chance to slow things down a bit and reset. Coach Matthew and I are going to have a lot of chocolate.

Carve Out

This week I read The Inner Game of Tennis.

Key insight: performance improves when you stop overthinking and let things happen. That applies just as much to running as it does to work.

You Brain Quits Before Your Legs

Trained cyclists who pedal to absolute exhaustion can immediately sprint at three times the power they just declared impossible — their muscles are fine, but their brain has decided the effort is no longer worth it. Samuele Marcora’s psychobiological model argues that perceived effort, not physiology, is the binding constraint on endurance performance, and the lab evidence is striking: a 90-minute cognitive task cuts time-to-exhaustion by 15% without changing a single physiological marker, while a two-week self-talk intervention improves it by 18%. The complication is that almost none of this has been validated in actual ultra-distance racing, where rhabdomyolysis and GI failure are real and no amount of mental reframing will override them. The piece breaks down where the science is strong, where it falls apart, and what specific psychological skills are worth adding to your training plan.

Full article here.

With that behind us, structure is starting to return. I’ve begun running to the office, which is eighteen kilometres door to door. For now, I’m splitting it by running in the morning and then picking up the Tube about fifteen kilometres in. It works, but it’s not the end state. The limiting factor is not fitness, it’s logistics. Running with a backpack carrying a full change of clothes adds unnecessary friction. The solution is straightforward. Once I’m fully set up in the office, I’ll leave clothes there and remove the constraint entirely. At that point, the full distance becomes viable and repeatable.

Training inevitably had to flex during the move. There was too much going on to maintain consistency at the level I expect, so weekly mileage came down. That was a deliberate adjustment rather than something to fight against. Now that we have stability, I’m back to around one hundred kilometres per week and building again towards one hundred and twenty. The approach is controlled, not rushed. In parallel, I’ve shifted nutrition. I lost weight over the past few months, so the focus now is on running a calorie surplus to support the increased load. More output requires more input, and there is no way around that.

At home, things are working. Alessia is settled, Matthew has a new routine and a new nanny, and he’s clearly comfortable in the new environment. That removes a significant amount of background friction and allows everything else to function properly. It also changes how the days feel. There is less noise, more clarity.

There are also moments that reinforce why this move matters. Running through Richmond Park early in the morning, with no one around and deer moving through the landscape, creates a completely different headspace. It’s quiet, controlled, and gives you room to think. In a city like London, that kind of environment is an advantage if you use it properly.

Alongside all of this, I’ve gone back into AI at a deeper level. Not just using tools, but building and running my own setup. I’m working directly in the terminal again, connecting research, data, and code into simple workflows that actually run. The impact is immediate. Things that used to take hours now take minutes, and iteration becomes much faster. You stop thinking in terms of tasks and start thinking in terms of systems. Instead of doing the work, you set up something that keeps doing it for you. That shift is the real unlock. It’s not about productivity, it’s about leverage.

The next phase is clear. Stabilise everything, remove remaining friction, and build volume back up. From there, it’s about getting back out into the countryside and lining up a few 50Ks. That’s where this goes.

Carve Out

This week I’m finishing Liftoff by Eric Berger, which focuses on the early years of SpaceX and the repeated failed attempts to get Falcon 1 into orbit before finally breaking through. The takeaway is simple. Progress comes from tight iteration loops. Build, test, fail, adjust, repeat. Over time, that compounds into something that looks inevitable from the outside.

I’ve also been documenting some of what I’m building and exploring on fdicostanzo.com, including a few live projects. One tracks recent UK house price transactions by postcode, and another compares F1 qualifying laps, like Verstappen versus Antonelli.

What follows is a thematic map of the year. No rankings. No scores. Each title includes a plot-accurate description of what it is fundamentally about.

Leadership, Power, and Decision-Making

These books examine leadership when outcomes are asymmetric, scrutiny is constant, and mistakes compound quietly.

Dear England — Gareth Southgate

Chronicles Southgate’s tenure as England manager and the cultural reset he engineered after years of underperformance. The narrative centres on psychological safety, accountability, and leading under national scrutiny.

The Ride of a Lifetime — Bob Iger

Follows Iger’s rise inside Disney and the decisions that reshaped the company through acquisitions, succession planning, and creative risk. The story is grounded in real boardroom trade-offs rather than abstract leadership theory.

Who Is Michael Ovitz? — Michael Ovitz

An insider account of building Creative Artists Agency and redefining Hollywood deal-making. The plot traces how influence is accumulated, defended, and sometimes lost at the very top.

DisneyWar — James B. Stewart

Reconstructs the internal battles during Michael Eisner’s reign at Disney. At its core, it is a story about governance failure, succession paralysis, and unchecked ego.

Creativity, Inc. — Ed Catmull

Explains how Pixar designed an organisation capable of scaling without killing creativity. The narrative focuses on candour, feedback systems, and leadership humility.

How I Built This — Guy Raz

A collection of founder journeys that emphasise struggle over mythology. Each chapter follows the messy, non-linear reality of building something durable.

Banking On It — Anne Boden

The founder’s account of building Starling Bank from scratch. The plot focuses on decision-making under regulatory pressure, capital constraints, and prolonged scepticism from incumbents and investors.

Technology, Innovation, and Acceleration

These books are about speed: how fast technology evolves and how slowly institutions and governance catch up.

In the Plex — Steven Levy

Documents Google’s evolution from a Stanford research project into a global platform. The plot centres on engineering culture, data-driven decisions, and the internal tensions created by scale.

Genius Makers — Cade Metz

Follows the deep-learning revolution through rival researchers, labs, and tech giants. The narrative tracks how academic breakthroughs consolidated into corporate power.

The Thinking Machine — Stephen Witt

A reported history of Nvidia’s rise and how Jensen Huang positioned the company at the centre of the AI economy. The plot follows long-term bets, internal culture, and the paranoia required to stay ahead of exponential demand.

The Optimist — Keach Hagey

A biography of Sam Altman and the ascent of OpenAI, framed around ambition, governance tension, and moral trade-offs. The narrative exposes the fragility of “mission-driven” organisations at scale.

Breakneck — Dan Wang

Examines technological acceleration through a geopolitical lens, contrasting Western inertia with Chinese execution speed. The book is built around urgency, institutional complacency, and strategic drift.

Source Code — Bill Gates

Revisits Gates’ early life and Microsoft’s formative years. The focus is on learning velocity, systems thinking, and how software reshaped economic power.

Corporate Failure, Risk, and Hubris

These books show that collapse is rarely sudden; it is usually rehearsed.

The Smartest Guys in the Room — Bethany McLean & Peter Elkind

An autopsy of Enron’s rise and implosion. The plot exposes how complexity, incentives, and arrogance masked fraud in plain sight.

Meltdown — Duncan Mavin

Traces the long institutional decay of Credit Suisse. The narrative follows repeated scandals, cultural erosion, and missed opportunities to course-correct.

Rogue Trader — Nick Leeson

A first-person account of how unchecked authority and weak controls destroyed Barings Bank. A study in operational blind spots and systemic failure.

Built on a Lie — Owen Walker

Investigates Neil Woodford’s rise as a star fund manager and the collapse that trapped retail investors. The plot focuses on liquidity mismatch, regulatory gaps, and reputation overpowering evidence.

Empire of Pain — Patrick Radden Keefe

Chronicles the Sackler family’s role in the opioid crisis. The narrative shows how wealth, legal strategy, and influence insulated accountability.

The House of Gucci — Sara Gay Forden

Follows the Gucci family’s internal power struggles over generations. Personal rivalries steadily dismantle a global luxury empire.

Hubris Maximus — Faiz Siddiqui

A critical examination of Elon Musk’s leadership, with particular focus on Tesla’s ascent. The story explores how celebrity, control, and constant escalation reshape organisational risk.

Boundless — Nick Kostov

The rise, fall, arrest, and escape of Carlos Ghosn, from corporate saviour to global fugitive. A boardroom thriller about power concentration, governance failure, and geopolitical consequence.

Endurance, Performance, and the Long Game

These books reject shortcuts and focus on accumulation over time.

In It for the Long Run — Damien Hall

Follows Hall’s ultrarunning career with emphasis on sustainability rather than extremes. Performance is framed as something to preserve.

Touching the Void — Joe Simpson

Recounts a catastrophic mountaineering accident in the Andes. A minute-by-minute study of decision-making when survival is uncertain.

How to Build a Car — Adrian Newey

Blends autobiography with engineering philosophy across Newey’s Formula One career. The story revolves around marginal gains, focus, and relentless iteration.

Inside Mercedes F1 — Matt Whyman

An inside account of Mercedes’ era of dominance in Formula One. Aligned leadership, systems, and culture sustain performance.

Total Competition — Ross Brawn

Tracks Brawn’s career across multiple championship teams. Preparation, adaptability, and execution are the recurring themes.

Money, Incentives, and Behaviour

These books explore how money reshapes judgement, priorities, and identity.

The Art of Spending Money — Morgan Housel

Explores spending as an expression of values rather than optimisation. Wealth is framed as optionality, not status.

Closing Thought

Across leadership, technology, endurance, and collapse, the same pattern repeats: systems fail long before outcomes do. Culture, incentives, and patience quietly determine results while attention is elsewhere.

That is the lesson this year’s reading reinforced—again and again.

Then the parent health tax arrived, right on schedule. Coach Matthew kicked off his seasonal immune system training block, and shortly after Alessia followed. I briefly convinced myself that good habits, sleep discipline, supplements, and borderline obsessive hand washing might keep me out of the blast radius. That illusion lasted until around 2am, when a toddler decided that sneezing directly into my face was the optimal delivery mechanism.

At that point, the outcome was inevitable. Training stopped. Three days out of play. Plans paused. This used to frustrate me more than it should. Now it lands differently. I have learned that fitness is not about never getting sick. It is about how quickly you bounce back when you do. Durability over bravado. Resilience over optics. The ability to absorb disruption without compounding it through bad decisions. I am saying this quietly and touching wood as I write, but the recovery curve is undeniably shorter than it used to be.

Forced rest created space. Instead of fighting it, I leaned into it. More time on the floor with Coach Matthew and his wooden railway. Frequent derailments. Tunnel explosions. Trains stolen and reclaimed through aggressive pursuit around the house. Chaos, but oddly restorative if you stop trying to control it.

This week reinforced something I keep relearning: sustainability beats intensity. In training, in work, in life. Taking a few days off did not erase months of consistency. It protected it. Just like stepping back during Christmas does not undo a year of effort, unless you decide that it does. This is the one period where excess is part of the design. Food, family, slower mornings. That is not failure. That is context.

The mistake is dragging guilt into January. The better play is to enjoy the pause, refill the tank, and restart with intent. Knowing when to push and when to stop is a skill, not a weakness. Progress comes from respecting limits, not pretending they do not exist.

Carve Outs

I have been watching The Last of Us on Vision Pro and it is absolutely phenomenal. The pacing, atmosphere, and emotional weight are exceptional. Pedro Pascal is outstanding, even if he looks slightly too similar to a banker I work with, which occasionally breaks immersion in a very specific way.

Short-form content finally did its job. YouTube Shorts convinced me to buy the Two and a Half Men box set. If you consistently enjoy thirty-second clips of something, there is a clear signal there. Either commit properly or move on. I chose to commit.

On the reading front, I finished The Art of Spending Money by Morgan Housel. As always, this is not really a book about money. It is about trade-offs, optionality, and defining what “enough” actually means. The core idea is simple and uncomfortable: the goal is not maximization, it is sustainability. Spend energy deliberately. Leave margin. Avoid decisions that feel optimal today but reduce flexibility tomorrow. I am now deep into Breakneck by Dan Wang, and a Book of the Year list will probably land soon.

Finally, my LEGO experiment has officially turned into an obsession. I am deep in the Technic rabbit hole. I built the Ferrari SF-24, a stunning set that is far more mentally demanding than it looks. Thirteen hundred pieces and over three hundred pages of instructions require full focus. I have also built Nigel Mansell’s Williams and Ayrton Senna’s McLaren. This hobby pulls me away from screens in a completely different way than running does. The only real constraint emerging now is space!

I didn’t buy much new gear this year. That, in itself, tells you most of what you need to know.

Footwear

Footwear was the most meaningful shift. Over the summer I transitioned to barefoot running, and it fundamentally changed how I think about shoes. The Merrell Vapor Glove became my default. Minimal, easy to carry, and honest in the feedback it provides, it worked across more terrains than I expected and forced me to move better rather than hide behind cushioning. That alone made it the most impactful piece of gear this year. When conditions turned muddy or technical, the NNormal Kjerag filled the gap. It delivered confidence and grip without feeling overbuilt, and unusually for a trail shoe, it actually looks good. Outside of running, I leaned heavily on Vivobarefoot for daily wear, extending the same barefoot principles into everyday life. Consistency matters, and switching biomechanics at the office makes little sense. On the racing side, the Nike Vaporfly remains my benchmark. I am a couple of versions behind and I did not race much this year, but when I do, that is still the shoe I trust. Proven still beats new.

Apparel

Apparel followed the same pattern. This was a year of using what I already had and realising how good it still was. Many of my most-used running tops are event T-shirts from past races. They are broken in, predictable, and comfortable in a way brand-new kit rarely is. I relied heavily on Runderwear garments for the top layer, which continue to do exactly what they promise with no fuss. For bottoms, Nike running pants bought last year or earlier are still going strong, showing no meaningful degradation despite heavy use. The same goes for Nike Trail shorts with pockets, which have quietly become some of the most practical pieces I own. They hold what they need to hold, they do not bounce, and they have survived repeated abuse. Nothing here is new. That is the point.

Hardware

When it comes to wearables and hardware, role clarity has been key. For daily life, the Apple Watch is unmatched. Nothing integrates with the iPhone ecosystem as seamlessly, and for notifications, convenience, it is exactly what I want. For training, however, I still rely on a Garmin Enduro from 2021. Four years on, it remains an exceptional endurance tool: reliable, durable, and with battery life that makes newer models feel unnecessary. Heart rate data comes from a Polar Verity armband, which I find far more comfortable than a chest strap and accurate enough for anything that matters. Indoor training is anchored by long-term investments that continue to pay off: a Wattbike Atom bought in 2019 and a Technogym MyRun treadmill from 2021, both still going strong and proving that good hardware does not age quickly. For audio, Shokz headphones are non-negotiable outdoors for safety and awareness, while an old pair of Bose Soundsport continues to serve me well indoors. Again, the pattern is clear: buy well, then stop thinking about it.

Apps

Apps have been about execution, not experimentation. I went back to Strava primarily for its maps and route builder, which remain best in class. TrainingPeaks continues to be the backbone of my training planning and syncs cleanly with my Garmin, doing exactly what it is meant to do. Zwift has been excellent for indoor cycling, particularly with ERG mode and structured workouts that remove decision-making entirely. For audio and entertainment, Audible, Apple Podcasts, and Apple Music have covered almost everything on the move, while Apple TV and YouTube have been invaluable for long indoor bike sessions where mental endurance matters as much as physical. ChatGPT has also become a genuine thinking tool for me, useful not just for planning but for challenging assumptions and reordering thoughts. On the recovery-data side, I used Whoop consistently and appreciated the passive data collection and low charging friction, even if I remain sceptical about the scores themselves. In principle, I still believe HRV4Training is the gold standard, but with a toddler in the house, consistency on waking measurements is hard, and data that is not repeatable quickly loses value.

Recovery

Recovery and injury prevention were about removing excuses rather than chasing optimisation. A Theragun Mini proved invaluable on the road, compact enough to travel everywhere and effective enough to actually get used. Resistance bands and a simple set of dumbbells kept me honest with strength training throughout the year. Nothing fancy, nothing complex, just enough load and structure to maintain durability. Availability mattered more than sophistication. If it was always there, it got used.

Nutrition

Nutrition followed the same pragmatic logic. For training fuel, I relied mostly on GU gels for their variety of flavours, which helps with compliance over long efforts, and on Maurten and Precision Fuel and Hydration when I wanted absolute consistency in carbohydrate intake. Supplements were deliberately boring: a multivitamin in a large glass of water on waking, vitamin D3, fish oil, and creatine. Protein supplementation with whey only came into play when travel made proper meals difficult. Otherwise, I stuck to whole foods: plenty of vegetables, especially broccoli and green beans, solid protein from salmon and chicken, healthy fats, and carbohydrates from fruit and bread. During heavy travel, Huel was a useful fallback when cooking simply was not an option. Functional, not aspirational.

Travel

Travel and work-life gear produced some of the quietest wins. A STNKY washable laundry bag turned out to be an unglamorous but brilliant purchase, keeping filthy running kit contained when moving between hotels. My North Face Fuse Box came on every trip this year and remains an absolute beast of a backpack: structured, durable, and utterly dependable. AirPods Pro were perfect for ease of access and sound quality during constant transitions. For downtime, the Steam Deck made travel significantly more enjoyable, offering real gaming without compromise. And then there is the Apple Vision Pro. For me, it has been a genuine game changer. The Mac virtual display dramatically improved productivity on the road, and cinema mode transformed movie watching in tiny hotel rooms. It reclaimed space, focus, and immersion where none exists.

Wildcard

The wildcard this year were the Insta360 Go and Ultra. Small, almost forgettable cameras that turned out to be perfect for capturing short bursts while travelling or documenting running adventures. They lowered the barrier to recording moments that would otherwise be missed, and that alone earned them a place.

The bad

Not everything survived. I tried Athletic Greens and Huel Daily Greens and abandoned both. The taste was poor, the pricing aggressive, and the value proposition weak. Everything they promise can be achieved more cheaply and more enjoyably through whole foods and targeted supplements. Popularity does not equal necessity, and convenience alone is not enough when the experience fails.

The game changer

If I had to buy one thing again tomorrow, knowing what I know now, it would be the Vision Pro. It is expensive, difficult to explain, and easy to dismiss on paper. But it is also one of the few devices in years that genuinely creates new behaviour. It feels magical in a way that only makes sense once you put it on your face. You cannot evaluate it intellectually. You have to experience it.

Looking back, this was not a year of marginal gains or constant upgrades. It was a year of letting good gear prove itself through repetition. Judgment over novelty. Execution over optimisation. That is the kit that earned its place.