Found this — WhatCable, a tiny macOS menu bar app that tells you what a USB-C cable can actually do. Speed, wattage, why your charging is slow, whether the cable (and not the charger) is the bottleneck.

https://github.com/darrylmorley/whatcable

Hi -

As usual: a few things I picked up in my peregrinations. This month: a marketing legend’s best lines, seven AI reports worth reading, two tools worth installing, some tangents that get progressively more unhinged, and a new section tracking conferences worth attending. Let’s go.

📚 READING

A break from the classics¹ this time. Instead, some great marketing advice from a legend in that space, and then a joke from someone that went to space.

Alchemy, by Rory Sutherland² — Sutherland is Vice Chairman of Ogilvy, and Alchemy is his case against the thing his own industry worships: data-driven, logic-first decision-making. His argument, roughly: once everyone in a market is being rational, rationality stops being a source of advantage — the value has already been arbitraged away, so the only edge left is doing something a spreadsheet would never approve. His favorite line, borrowed and repeated in every talk he’s ever given: value isn’t created in the factory, it’s created in the mind — a $4 bottle of water and a $0.40 one are chemically identical, the entire delta is psychological. Not a book I’d necessarily recommend cover to cover (it repeats itself, and some of the “11 rules of alchemy” framing is padding), but he’s a great talker, so definitely listen to him instead for example on the “Killer Marketing Secrets” episode and you’ll walk away with better lines than the book gives you. A few from the book that stuck anyway:

· A flower is simply a weed with an advertising budget.

· The problem with logic is that it kills off magic.

· Solving problems using rationality is like playing golf with only one club.

· The human mind does not run on logic any more than a horse runs on petrol.

· If you are wholly predictable, people learn to hack you.

· The fatal issue is that logic always gets you exactly the same place as your competitors.

· “The heart has reasons of which reason knows nothing”, as Pascal puts it. [I love this one – so true!]

· Human nature hasn’t changed for a million years. It won’t even change in the next million years. Only superficial things have changed.

· Humans may be descended from ancestors who were better at concealing their true motives. It’s not enough to conceal them from others — to be really convincing you also have to conceal them from yourself.

· It is difficult to get a man to understand something when his salary depends on his not understanding it. — Upton Sinclair

· “A change in perspective is worth 80 IQ points,” so said Alan Kay.

· Advice for young people: find one or two things your boss is rubbish at and be quite good at them.

· Everyone likes to go to a nightclub in the company of a friend who’s slightly less attractive than them.

· Advertising: the word comes from the Latin “anima advertere,” or “to direct attention.”

· The psychophysicist Mark Changizi has a simple evolutionary explanation for why water doesn’t taste of anything: the human taste mechanism has been calibrated not to notice water’s taste, so it stays optimally attuned to the taste of anything that might be polluting it.

· If rationality were valuable in evolutionary terms, accountants would be sexy. Male strippers dress as firemen, not accountants. Bravery is sexy; rationality isn’t.

· Conventional logic is hopeless in marketing — you end up in the same place as your competitors.

A joke from the early 1960s: After Yuri Gagarin, the first cosmonaut, made his visit to space, he was received by Nikita Khrushchev, the general secretary of the Communist Party, who told him confidentially: “You know, comrade, up there in the sky, I saw heaven with God and angels — Christianity is right!” Khrushchev whispered back: “I know, I know, but keep quiet, don’t tell this to anyone!” The following week, Gagarin visited the Vatican and confided to the pope: “Holy father, I was up there and saw there is no God or angels...” “I know, I know,” the pope interrupted, “but keep quiet, don’t tell this to anyone!” [I guess you could play the same game today, but instead of asking about the existence of God, it’s the existence of ROI from AI! 🤣]

🤖 TECH

Seven reports this time, still mostly AI, but for once with more signal than noise.

Almost all the AI value went to the picks-and-shovels, not the apps³ — Jefferies tallies ~$22T of market cap created since ChatGPT launched, with infrastructure (compute plus silicon) capturing about two-thirds of it (page 3). Memory providers are up over 4,000% since launch while application software is actually down (page 4). On the private side, foundational-model startups hoovered up ~70% of top-50 VC dollars raised this year versus ~2% for the application layer (page 18). The one that stuck with me: AI-native firms run at roughly 8x the revenue-per-employee of traditional SaaS (page 12). Worth a flip.

Enterprises love AI pilots and can’t ship them⁴ — Madrona surveyed 150 enterprise decision-makers: about three-quarters plan to expand AI budgets, but roughly 80% convert fewer than half their pilots to production (page 3). Surprisingly, “didn’t work as promised” is only the sixth most common reason a pilot dies — integration and security top the list (page 3). There’s a pricing disconnect too: a third of buyers want outcome-based pricing, but hardly anyone gets it (page 14). On jobs, they’re blunt: most reshaping shows up as slower hiring, not layoffs (page 4).

AI adopters are hiring more, not less⁵ — Ramp’s own data quietly detonates the “AI is coming for the jobs” narrative: firms adopting AI heavily grew headcount about 10% over two years, entry-level even more, while low-intensity adopters looked no different from the control group (pages 6-7). Contract values for AI software have roughly grown 10x in two years, headed toward ~$1M by 2026, per Ramp’s projections (page 12). Anthropic has now edged past OpenAI in business adoption (pages 2, 10). Still a rounding error next to payroll though — AI spend per employee is a fraction of one month’s salary (page 3).

Canaries in the coal mine: AI is hitting young workers⁶ — Brynjolfsson, Chandar and Chen (Stanford)’s updated ADP study shows no economy-wide displacement — overall employment is up since late 2022 (page 8). But young workers (22-25) in the most AI-exposed jobs are now about a fifth below where they’d otherwise be, and the gap keeps widening (pages 8, 10). It runs through hiring, not layoffs — separations haven’t spiked in exposed jobs (page 16). The effect flips positive for the 41+ crowd, who benefit from AI as a complement rather than a substitute (page 20).

Everyone found ROI religion, nobody actually cut the AI budget⁷ — UBS’s Karl Keirstead worked the Ai4 conference for unvarnished takes from ~15 enterprise IT execs (page 1). The mood has shifted from “token-maxxing” to “token optimization,” but none of that discipline has actually dented spend yet (page 2). Appetite for cheap Chinese models stays low on security concerns (pages 3, 9). Build-over-buy is still alive — one F500 exec is rebuilding their own Palantir alternative because FDEs bill $650/hour (pages 4, 7). Dense, quote-heavy. Good to hear directly from buyers.

Leaders and workers are describing two different companies⁸ — Notion surveyed over 6,000 professionals: about 90% of orgs are stuck using AI as a personal productivity tool rather than running real workflows (pages 4, 7). Leaders think their org is five times further along than employees say it is, on the same question (page 7). Leaders are also about twice as confident in their org’s AI capability as the people actually using it (pages 4, 9). Seniority is the strongest predictor of who believes the hype — CEOs are roughly six times more likely than individual contributors to say their org has arrived (page 16).

The gains don’t spread, they concentrate: the top engineers ship roughly 50x the median’s AI-written code⁹ — Bessemer interviewed CTOs at 20+ companies and found real inequality, not acceleration — the best engineers ship roughly 50x the AI-written code of the median (pages 5, 14). Real agentic engineering isn’t cheap — it runs at least $1K/month per engineer (page 5). Uber rolled Claude Code out to thousands of engineers and burned its entire annual AI budget in about four months (page 14). Salesforce is spending hundreds of millions on tokens while freezing engineering headcount — openly swapping tokens for people (page 14).

🔧 TOOLS

Two this time: One to make you a wiz at presenting, the second that allows you reduce your LLM bills.

ZoomIt for Mac¹⁰ — Microsoft finally ported the legendary Sysinternals ZoomIt to macOS: a menu-bar utility for live/static zoom, drawing and typing annotations, region screenshots with OCR, MP4 screen recording with optional webcam overlay, and scrolling panorama capture. Install via brew install --cask zoomit, then Control+1 to zoom, Control+5 to record. Easy must-install if you present or record demos.

CanIRun.ai¹¹ — a searchable catalog for figuring out which open-source AI models your actual hardware can run, graded against your GPU or Mac specs, covering chat, coding, image/video, and lightweight models for constrained devices. As more people move to local models, this may be handy.

🎲 TANGENTS

Five quick ones — a two-thousand-year argument for funding curiosity, a couple of funny videos, a peek under PowerPoint’s hood, and a story of a man trying to save coffee from becoming a luxury.

The case for funding things nobody can justify yet¹² — Two things that pair well: Abraham Flexner’s 1939 essay The Usefulness of Useless Knowledge, and an awesome video from Dr. Roy Casagranda on the Great Library of Alexandria. Flexner’s argument: wireless communication came from Maxwell’s and Hertz’s abstract 1870s equations on electromagnetism, done with zero interest in application. Non-Euclidean geometry sat unpublished for 25 years before it became the mathematical backbone of relativity. Flexner used the essay to justify the Institute for Advanced Study’s charter, where professors get “no duties, only opportunities” — a policy that let it scoop up Einstein, von Neumann, and Weyl the moment they fled Hitler’s Germany. [Given that we have our own dictator now, not surprising that the US is losing scientists to other countries where once it was a beacon of science. Amazing how much damage one 🍊idiot can do.]. Dr. Roy’s video is the same lesson run in reverse: Rome burned the Great Library and Emperor Theodosius purged non-Christian scholarship from the empire, and the knowledge that survived only did because scholars smuggled their books out to Persia’s Academy of Gundishapur — which is where the Islamic Golden Age picked up the thread centuries later. Dr. Roy’s scholarship never ceases to amaze me.

Vibe coding, hilarious¹³ — If there’s one video that captures the promise and the angst of vibe coding, this is it. It’s based on a famous “meme” video of Hitler berating his generals and videos in German, but the sub-titles in English are what make this a banger. Please just go and watch. The best ~4 min you’ll spend all week. The creator deserves an Academy Award + Palme d’Or + BAFTA!

European tech, unstoppable¹⁴ —”How Americans see EU Tech” — a deadpan run through American stereotypes about European tech being behind: 3G towers, floppy disks, “5kb transformer models,” checking GDPR notices at a mailbox, a protest as the “national pastime.” It wouldn’t be so funny if it weren’t so true. (Hat tip to Neal for pointing this out to me. 🙏). And if any Europeans want a rebuttal, wasn’t it the Greeks that created the first analog computer – the Antikythera mechanism, even though that was ~2000 years ago.

What’s inside a PowerPoint file¹⁵ — turns out a .pptx is just a ZIP archive full of XML you could reconstruct by hand. The spec that governs all that XML runs past 5,000 pages. Coordinates inside the XML are measured in EMUs (English Metric Units, 360000 EMUs = 1 cm) instead of inches, purely to avoid floating-point math, so every shape’s position ends up described in absurdly large integers. One theory is that Microsoft made the format deliberately painful to reverse-engineer; the less conspiratorial read is that any 40-year-old product accumulates this much compatibility debt on its own. Either way, it’s exactly why automating PowerPoint edits with code — or with an AI agent — is so much harder than it looks. Great example of a company using clear explanations to signal their knowledge of a space.

The maverick scientist trying to save coffee from extinction¹⁶ — The global coffee industry spends less than 0.1% of its ~$250bn in annual revenue on R&D, even though 2.25bn cups get drunk every day and half the world’s coffee-growing land could be unusable by 2050. Climate change has already more than doubled the US price of ground-roast coffee since 2020, and the prediction is blunt: “Within a decade, coffee will become a luxury product.”. The first thought that came to mind is that is awful for all my coffee drinking friends. The second (inside voice) thought that came is thank God I drink chai.

📅 EVENTS ^(new!)

New section this month on upcoming events. Normally I’m chained to my laptop in a dungeon. Occasionally, they let me out and mingle but to minimize the potential (reputational, brand, limb) damage, they limit my exposure to small private dinners, and 1:1 meetings with enterprise companies, startups, VCs etc. On rare occasions, I’m let out into the public arena without a muzzle (sort of like Christians at Carnival before Lent). A few of these public events (and one private one) happen to fall in September.

· Goldman Sachs Technology Conference — Sep 8–11, 2026, San Francisco. Agenda / event site. Chock full as always with the tech heavy weight champions of the world (think Jensen, Bill, Michael, Dara, etc.). Guaranteed to have memorable quotes emerge.

· All In Summit — Sep 13–15, 2026, Los Angeles. theallinsummit.com. This is not my usual cup of chai. Last time I went I think I was one of the only people that didn’t listen to their propaganda podcast religiously. I know, blasphemy. To be fair, they do get the “who’s who” as guests.

· Beyond “Off the Rack” AI — September 17, 2026, San Francisco, private dinner for ~8-10 CxOs on the topic of open weights models, post-training and sovereignty. Interest in this one has been unusually high — we’re close to full already. But if you have any CxOs that may find it useful, let me know.

· Madrona IA40 Summit — Sep 30, 2026, Seattle. ia40.com/summit. The panels are usually ok / not great, but they have a good mix of VCs and startups. And the focus on AI apps helps.

If you are going – would love to meet. They don’t let me out of the cage that often…

[By the way, if someone knows some really authentic / kickass (karak) chai places in either Los Angles or Seattle, please let me know. Asking for a friend. My friend already knows a great place in downtown San Francisco but maybe going to the events above, hence his ask! 🤣]

A few more that didn’t make the cut but may be worth a skim: (i) Guggenheim, Monthly Software Market Update¹⁷ — application software now trades at just ~3x forward revenue, less than half its ~7x five-year average, while median net dollar retention slid from ~115% in 2021 to ~110% in Q1 2026. (ii) Jefferies, Quarterly Software Earnings Summary¹⁸ — ~85% of the ~140 software firms beat Q2 CY2026 revenue consensus, down from ~90% the prior quarter, on median YoY growth of just ~15%. (iii) Jefferies, Software Market Valuation & Performance Update¹⁹ — “AI Darlings” are up ~500% since ChatGPT launched while horizontal application software is down 3%, and only four names — Broadcom, DigitalOcean, Oracle, Palantir — still grow 30%+ NTM. All saved in the “cutting floor” folder.

Sources: ¹It actually takes time and effort to read these classics. They are good, but hard to get through! ²Rory Sutherland, Alchemy: The Dark Art and Curious Science of Creating Magic in Brands, Business, and Life (Amazon). ³Jefferies Technology Investment Banking, GenAI Industry Insights, Valuation, and Performance Update (Q3 2026). ⁴Madrona, Harnessing Enterprise Value – The ROI of AI (August 2026). ⁵Ramp (Ramp Economics Lab), The Ramp AI Index — A Compiled Look at U.S. Business AI Adoption (compiled June 2026). ⁶Brynjolfsson, Chandar & Chen (Stanford Digital Economy Lab), Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence (August 2026). ⁷UBS Global Research (Karl Keirstead & Jack Fyda), AI Research: Latest Enterprise AI Checks (August 10, 2026). ⁸Notion, The Great Renovation: The State of Global AI Transformation (2026). ⁹Bessemer Venture Partners (Liran Eshel & Adam Fisher), The Agentic Awakening (June 2026). ¹⁰ZoomIt for Mac, GitHub (github.com/microsoft/ZoomitForMac). ¹¹CanIRun.ai (canirun.ai/models). ¹²Abraham Flexner, The Usefulness of Useless Knowledge, Harper’s Magazine (June/November 1939); Dr. Roy Casagranda, The Great Library of Alexandria, Museum of the Future: Lessons from the Past (youtube.com/watch?v=OyPbCYnuFa8). ¹³Paul Copplestone (@kiwicopple) on X (x.com/kiwicopple/status/2089917700822999275). ¹⁴Neal Zuckerman via Matthias Schmidt (@eurofounder) on X (x.com/eurofounder/status/2091944123838108152). ¹⁵Editide, What Is a .pptx File? (editide.com/blog/what-is-a-pptx-file). ¹⁶Tomas Weber, The Economist 1843, The Maverick Scientist Trying to Save Coffee from Extinction (July 23, 2026) (economist.com/1843). ¹⁷Guggenheim Securities, Monthly Software Market Update (cutting floor). ¹⁸Jefferies, Quarterly Software Earnings Summary (cutting floor). ¹⁹Jefferies, Monthly Software Market Valuation and Performance Update (cutting floor).

Hope this helps.

(For those wondering, yes this is human created – with tons of human labor and love, and a little help from AI. And no, I don’t have anyone helping me apart from the few kind souls – thank you! – that send along things I should consider.)

Thanks,

Akash

p.s.: If you are looking for back issues of this, they can be found here.

Ubs Latest Enterprise Ai Checks

250KB ∙ PDF file

Download

Download

Madrona Research Harnessing Enterprise Value Roi Of Ai

681KB ∙ PDF file

Download

Download

The Agentic Awakening

649KB ∙ PDF file

Download

Download

The Usefulness Of Useless Knowledge Flexner 1939

2.78MB ∙ PDF file

Download

Download

Ramp Ai Index

3.08MB ∙ PDF file

Download

Download

Canaries August2026

3.23MB ∙ PDF file

Download

Download

Jefferies Genai Industry Update (q3 2026) Vf

2.02MB ∙ PDF file

Download

Download

The Maverick Scientist Trying To Save Coffee From Extinction The Economist

5.65MB ∙ PDF file

Download

Download

Jefferies – Quarterly Software Earnings Materials August 2026 Vf

268KB ∙ PDF file

Download

Download

Jefferies – Software Valuation Update (august 2026) Vf

1.14MB ∙ PDF file

Download

Download

Guggenheim+ +software+market+update+july+2026

982KB ∙ PDF file

Download

Download

Researchreport Thegreatrenovation

18.5MB ∙ PDF file

Download

Download

If anyone is wondering if the software security situation is going to get worse with AI, exhibit A below, from a colleague who is non-technical.

In his words (redacted the tool name):

"In the illicit trade of automation scripts to save sanity... a script to turn off redacted for 10 minutes with one click (goes back on automatically). Built it, with Codex. I now have it as a dock button (Apple Shortcuts app)... I was so fed up of reading that my agent got blocked and spent 10 minutes trying to reach a website because of redacted!"

Anyone else seeing this?

So friggin' true, all AI websites look the same bro!

op.tngl.io/every-fucking-website

I had a chance to spend a (short) bit of time at the Jefferies conference on tech for CFOs in NYC last week.

Thanks to Evan Osheroff for having me and hosting. As you can imagine, the questions were all about AI - I know that's a shocker!

A couple of the themes that came up - enterprise adoption, cost, incumbents:

• On enterprise adoption: I see adoption as actually pretty low. You of course see individual productivity (Bobby is 10-20% better because of Claude), but not sure if his entire function (say finance) is better because of it. You certainly don't see the velocity / throughput accelerate for all of the company. (The AI native companies are of course building these company wide "operating systems" - https://www.linkedin.com/feed/update/urn:li:activity:7492958608532525056/).

• On the rise of open weight models, and the desire for model independence, cost relief and taking back control: As usual, the AI native companies (e.g. Cursor) have been way ahead of this for a while and having been building this model flexibility into their products. Most (non tech) enterprises are way behind this but I think this is going to bubble up imminently -- the LLM bills are adding up while the actual business impact is still more hand wavy.

• On incumbents: Everyone has written them off, but I think one thing that all the hordes of AI native startups are ignoring is that the incumbents are already "in the building & connected". No CIO/CISO is going to open up the kimono to all their internal systems to dozens & dozens of startups. Consolidation will happen!

(Btw, it was pretty beautiful inside, while being a fortress outside).

Thanks to Jake Saper and Ryan for hosting us at a great AINS event in NYC last week. It was great -- full of budding founders, people curious about AINS and some established companies.

I liked the format -- they first had three AINS (Crosby, Hanover Park, Pace) startups, and then an established SW company (Gainsight) talk about their experiences, followed by a group discussion. A couple of tidbits:

• They had room for 75, let in 100. The wait list was for 400 people! (This is how cults start).

• Usually services business are ~30% gross margin, AINS are ~70% (~software margins). Hence the interest from VCs.

• "You have to build BCG and Stripe simultaneously." i.e. not easy at all.

• On incumbents building this internally vs. starting from scratch, one of the topics that came up is the investment required. E.g. Kirkland & Ellis (the law firm) is spending 1% of their revenue on R&D vs AINS companies spending a lot more (I'm assuming akin to SW companies). What was not said is also the ability for incumbents to attract and retain software talent -- I don't think that's easy at all. Would a CS grad go to a law firm or an AI company?

• Gainsight is effectively three companies in one. The first two are classic SW/SaaS businesses. The third is AINS business -- renewals for the long tail in a pricing model that is a combination of a management fee and an outcome fee. "It's a TAM expansion business".

• "If a service goes wrong it's easier to cast blame on a human than AI". I'm assuming since people are more forgiving for (decent) humans.

• One of the AINS (serving insurance providers) said their offering used orders of magnitude fewer humans than the competition. Read that again -- orders of magnitude. I hope the BPOs are paying attention.

All in all pretty interesting. The part that left me wondering is why more software companies don't do what Gainsight is doing. Most of them are tinkering at the edges of changing their services, CS, renewals teams, but I don't know of any that have taken a bold step as has Gainsight to put their money where their mouth is and actually "run a service". I think it's a transformative move that more should seriously think about doing.

• "This reprehensible carelessness may be related to the pressure of circumstance, since we are war with France, a fact which may come as a bit of a surprise to you gentlemen in Whitehall."

• "I shall pursue either one with the best of my ability, but I cannot do both: 1.To train an army of uniformed British clerks in Spain for the benefit of the accountants and copy-boys in London or perchance. 2. To see to it that the forces of Napoleon are driven out of Spain."

It reminded me of all the bureaucratic bullshit that people that are on the front lines and actually do the work have to deal with vs. the back office monkeys / number crunchers who try their best to impede.

There's no wonder people get fed up and leave big companies.

https://wellsoc.org/society-member-pages/anecdotes-of-wellington/

Every AI-native company seems to be quietly building its own OS.

E.g. Cloudflare open-sourced Cloudflare OS on Aug 5 — an agent workspace grounded in curated company context, sandboxed apps employees build for themselves, and a permission layer called Gatekeepers.

They're not alone. Vercel runs 100+ internal agents behind a single Slack handle (@v), on a framework called eve they open-sourced in June. Sierra calls its internal agent Pinecone "the operating system for our company". Ramp has Glass. Stripe has Kai. Block open-sourced Goose. All these are somewhere between a Glean and an enterprise-grade OpenClaw or Hermes.

It's an interesting trend, since I strongly believe that companies that have such "OS"es that allow the whole company (not just individuals) to go faster will be in a killer position vs. those that are either just getting to Claude or heavens forbid calling it done after installing Microsoft CoPilot for everyone. (Almost everyone has individual productivity done; it's company wide productivity that's going be super hard to get right.)

(By the way, I'd be surprised if some of these don't become the main thing they sell. There's good precedent. Slack was an internal chat tool Tiny Speck built while making a browser game called Glitch. The game shut down in December 2012; the team looked at what they couldn't stop using and productized it. $27.7B to Salesforce nine years later.)

Cloudflare OS blog: blog.cloudflare.com/cloudflare-os/, github.com/cloudflare/cloudflare-os

Vercel eve: vercel.com/blog/introducing-eve · github.com/vercel/eve

Sierra on Pinecone: sierra.ai/blog

Block's Goose: github.com/block/goose

Stripe's Kai + Minions: stripe.dev/blog

The agentic close, autonomous AR, AI-driven FP&A: the Office of the CFO is quietly becoming one of the most contested categories in enterprise software. (I recall one company calling themselves the "Palantir for the CFO"!).

In that vein, I'll be at Jefferies' Office of the CFO Software Summit at the NYSE on Wednesday, August 12, doing a lunch fireside chat on the state of AI in enterprise. (Of course I'll try to make it spicy; the chat that is. I don't control the lunch.)

Twenty-plus companies on the agenda, each pitched as "the future of the OCFO." The interesting question: which of these stay standalone in three years, which get swallowed by incumbents like NetSuite, Workday, or SAP, and which get swallowed by the AI-native platforms trying to become the new incumbent themselves (Ramp, Campfire, Rillet)?

https://www.jefferies.com/events/office-of-the-cfo-software-summit/

AI-native services are moving from interesting concept to a transformational category.

I'll be at the AI-Native Services Summit in New York on Wednesday, August 12, joining founders and investors working through what it takes to build and back this model.

Who else will be there?

Thanks to Jake Saper for organizing and the invite, and to Ryan at Crosby for hosting.

https://www.linkedin.com/feed/update/urn:li:activity:7490130397931159552/

Hi -

As usual, I wanted to share a few things that I’ve found interesting in my aimless wanderings (story of my life!). In the mix tape this time: One of my favorite philosophers from 300 BC, a bunch of AI reports, some cool tools, and a few interesting detours. Let us dive in.

📚 READING

A couple of things this time, one from ~300 BC, one from a few weeks ago.

Diogenes¹⁹ — For those that don’t know, Diogenes was a Cynic philosopher (c. 404–323 BC) who “lived in a storage jar” in the Athens marketplace, owned nothing, and made it his life’s work to mock anyone who took wealth, status, or convention seriously – my kinda guy! Most of what survives comes down to us as anecdotes and one-liners (gathered in Diogenes the Cynic: Sayings and Anecdotes). He’s one of my favorites in addition to Epictetus. A few that stood out:

· When asked whether a certain man was wealthy, Diogenes replied that he had no idea, because he didn’t know how the man used his wealth.

· He said that, just as slaves are at the beck and call of their masters, so bad people are at the beck and call of their desires.

· When asked what is the right time to marry, he replied: “For those who are young, not yet; for those who are older, never at all.”. 🤣

· Leading away a man who had stolen from the temple treasury, he said: “The big thieves are arresting the little thief”

· The prosperity and good fortune of the wicked, Diogenes used to say, provides telling evidence against the power of the gods. (So true, e.g. see the punishment that the people behind the Epstein files have had to not suffer through).

· “If I was not Alexander, I should be Diogenes” — Alexander the Great. (This comes from their famous encounter where Diogenes told Alexander to get out of the way since he was blocking the sun!)

· When asked what is wretched in life, he said: “An old man without means”

· When asked what weighs most heavily on the earth, he said: “An uncultivated man.”. (I think ~8B people feel that currently given the rantings of the current occupant of the White House). See also the history of the last real-estate tycoon to try to invade Persia.

· Seeing an incompetent archer, he sat down beside the target, saying he wanted to make sure he didn’t get shot.

· He used to say a rich man who doesn’t know how to use his wealth doesn’t own it; it owns him.

· When asked who is subject to the greatest anxiety, he said: whoever aspires to the greatest prosperity and happiness.

· Seeing an envious man with a glum expression, he recalled what Bion the sophist said: either something bad has happened to him, or something good has happened to somebody else.

· He said teachers deserve praise from their pupils twice over: “From the gifted ones, for how much they learned, and from the incompetent ones, for how much trouble they caused.”

· Asked in what respect philosophers surpass other people, he replied that if all the laws were repealed, they would go on living exactly as they do now.

· To someone boasting he could drink large amounts without getting drunk, Diogenes said: “So can a mule.”

I think the Greeks had it figured out even way back then…we’re just trying to re-learn all that they already knew.

Rotten Tomatoes: Interesting quote from someone that didn’t like the recent Odyssey movie: “If I saw this on a plane, I’d still walk out”. 🤣

🤖 TECH

Seven reports this month, mostly AI (what else!): generally, a bunch of people trying to get out of the “frontier model taxation without ROIpresentation”™️. 🤣

ICONIQ survey of AI-company execs ³ — I’ve been a big fan of Iconiq reports, and this one doesn’t disappoint either. Anthropic’s share of the provider mix went from about half to over 80% in six months (page 10). R&D, sales, and product teams are set to grow headcount over the next two years while customer support and G&A shrink (page 26). Forward-deployed engineers have gone mainstream, about half of companies are scaling up FDEs as a permanent part of their go-to-market motion, mostly as a revenue driver rather than pure support (page 28), though nobody agrees on how to charge for it: roughly a third bundle it free into the subscription, another third charge it as a separate services fee, and the rest split across hybrid and deal-by-deal models (page 31). Internally, adoption is uneven by function too: R&D has AI tool access at 71% and a 3.7 maturity score, versus 59% and 54% access for G&A and GTM, both scoring around 3.1-3.3 (page 36). Put in the must look at category.

Goldman says the AI “job apocalypse” isn’t coming¹ — Goldman’s economists peg AI-driven job losses at around 9% of the workforce, roughly 15 million people, over a 10-year adoption cycle (page 10). MIT’s Daron Acemoglu expects white-collar “routine” roles to take the brunt of it; his colleague Neil Thompson calls it a “rising tide,” not a “crashing wave” (pages 4, 6). A couple of things that stood out: Industries with established AI use cases are seeing employment contract 11k a month (Page 8) – tech layoffs are roughly running at 100k this year alone, the US economy creates ~30m gross jobs per year (page 11) – note that it’s gross not net. Worth a flip, even though I don’t think anyone really knows.

Amplify Partners surveyed AI engineers, and ~90% say there are more productive² — About 60% evaluate what they’ve built through “manual review, vibe checks,” not tests or benchmarks (page 16). Hallucinations remain the top complaint at roughly 60%, yet about a quarter let their agents write to production unsupervised anyway (pages 13, 15). Engineers are bullish on themselves, nearly 90% say they’re more productive with AI, but only about a third think AI writing software is a solved problem (page 24). Two-thirds expect a lab to declare AGI within five years. Good data if you sell to dev teams.

Morgan Stanley put software through five AI stress tests⁸ — Debate 1: does all AI value accrue to the frontier labs? No, because LLMs have what MS calls a “Jagged Frontier”: brilliant at coding and generic knowledge work, unreliable at the messy specialized stuff like healthcare, legal, and finance, so application, infrastructure, and cybersecurity vendors who supply that context still win (page 95). Debate 2: does every app have to go “headless” for agents? Only if your moat was ever just the interface; Salesforce’s own pitch for its headless product is literally “why should you ever log into Salesforce again,” but vendors sitting on proprietary, regulated data can stay closed and keep the value (page 103). Debate 3: can vendors charge for anything beyond seats? MS expects a hybrid seat-plus-consumption model almost everywhere, with pure-consumption vendors like Snowflake, Datadog, and Shopify getting there faster and cleaner (page 107). Debate 4: do gross margins fall permanently once every workflow is burning tokens? Compression is real, but cheaper inference, caching, and model routing should hold EBIT roughly flat, the same way payments businesses learned to live with lower gross margins (page 118). Debate 5, “Tokenomics”: is AI pricing about to get audited the way cloud spend was post-Covid? Partly, mostly for experimental token burn rather than production use, and Google’s own price hikes (Gemini Pro’s output price roughly quintupled) show the “AI keeps getting cheaper” story is already more complicated than it looks (page 128). Their overall conviction, across all five debates: Microsoft, Palo Alto Networks, CrowdStrike, Shopify, Cloudflare, ServiceNow, Datadog, and Snowflake.

Mozilla’s case for open weights⁵ — With all the discussion on open models, thought it’d be helpful to share this report from Mozilla. A few tidbits: open models now trail the top closed ones by just a few points on Chatbot Arena, and the top five models by token volume on OpenRouter are all open weight (pages 5, 11). Open weight’s share of all OpenRouter tokens has climbed from about 2% in late 2024 to roughly a third by mid-2026, still climbing, largely on the back of Chinese models: Alibaba’s Qwen alone out-downloads the next eight organizations combined (pages 10, 31).

Piper Sandler polled CIOs⁶ —The barrier CIOs worry about most has quietly shifted: data quality now outranks security as the top obstacle to GenAI adoption, a real reversal after years of security dominating the conversation (page 34). The vendor pecking order is getting shaken hard: Microsoft’s share of CIOs calling it the “most strategic” AI vendor collapsed from 33% to 14% in six months, while Google took the top spot and Anthropic posted the biggest gain of anyone in the survey (pages 28, 57). Even Microsoft Copilot, still the single most popular AI app in the survey (really, have people actually tried to use it?), saw usage tick down alongside Salesforce Agentforce and ServiceNow’s AI agents; the only app gaining ground was Adobe Firefly (page 34). IT budgets overall are barely growing, but nearly all CIOs plan to spend more specifically on AI infrastructure. Lots to chew on in here.

Exponential View priced the AI economy at roughly $175bn⁴ — Azeem Azhar’s team built a bottom-up model deduplicating spend across apps, models, and hosting, and lands on an annualized run rate around $175bn, growing about three times faster than any prior IT platform shift (page 7). Inside the data center buildout itself, the mix has shifted hard toward silicon: chips now take about 60% of every dollar spent, up from 40%, and memory alone went from a 2% rounding error to nearly a fifth of spend (page 17). At the company level the top 10% of companies spend about $1000 per employee per month, whereas the median spend about $10 per user per month – that’s a log scale! (page 21). The reason spend keeps climbing anyway: an agentic coding task burns roughly 1,200 times the tokens of an ordinary multi-turn chat exchange (page 42). And the frontier labs are already losing share of that token pie: Google, OpenAI, and Anthropic together dropped from 72% to about 33% of weekly OpenRouter token volume, open-weight models are eating the rest (page 60). Really good report.

🔧 TOOLS

Two tools this time, both about taming the chaos on your screen.

Zen Mode¹⁷ — A script that gives macOS a real focus mode: hit Option+Cmd+Z and your active window glides to center screen and stretches to full height, while the Menu Bar, Dock, wallpaper, and every other window fade to black. Esc brings everything back exactly where it was. Works with any app, no per-app setup, because it operates at the window-manager level rather than hooking into individual programs, and it tracks the window live if you drag or resize it mid-session. Requires Hammerspoon.

Abralo¹⁸ — free desktop app (Mac, Windows, Linux, no signup beyond your existing Claude or OpenAI credentials) for running several coding agents at once without your terminal turning into soup. Claude Code and Codex sessions each get their own column, with token burn rate, weekly limits, and per-agent usage shown as sparklines, so you know which one is about to hit a wall before it does. If an agent needs your input, it nudges you instead of silently idling; if you run out of tokens mid-task, that turn gets held rather than lost, so you can pick it back up later instead of restarting.

🎲 TANGENTS

Four detours, none of them (thank God) about AI, all them worth a look.

Google trends circa 1933⁹ — Hoover convened a committee of economists and sociologists to produce a complete scan of the country, and in 1933 they delivered two volumes, 29 chapters, covering literally everything. The automobile alone rewired the country inside a decade: cars entering national parks went from ~14k (1916) to ~900k (1931), and the AAA estimated 45 million Americans took a vacation motor tour in 1929 alone (pages 469, 482). Medicine’s gains were oddly selective: infectious-disease deaths collapsed for the young, but mortality for adults over 50 kept climbing for as long as anyone had records, sparking a real debate over whether modern medicine was breeding a nation of longer-lived “weaklings” (pages 706-708, 714). And 24 states had compulsory sterilization laws for “defectives” on the books the whole time, blessed by the Supreme Court in 1927 over a single dissent (page 1560).

A history of American menus²⁰ — Someone did a solid and built a super elegant site based on thousands of restaurant menus from 1880–1920. Some tidbits: Delmonico’s opened in 1837, invented the American restaurant (individual tables, ordering off a menu instead of one set communal meal), and wrote the whole thing in French, imported chef included. A few blocks away, Smith & McNeil’s sat 1,000 people and fed roughly 10,000 a day, plain English, built for speed, not status. Café Savarin ran an all-French menu in 1888, and by 1915 was serving French, German, Italian, Hungarian, British, Indian, Japanese, and Caribbean dishes off the same card, riding the immigration wave. [Given the current trajectory, maybe we’ll all be forced to eat “hamberders” for patriotic reasons]. American dining started as rich people cosplaying French aristocracy, and only became genuinely American once the middle class forced its way onto the menu.

Richard Hamming, on why most careers wander²¹ — Clay Bavor (co-founder at Sierra) tells every new grad to read Hamming’s 1986 talk “You and Your Research,” calling it maybe the best return on words of anything out there. The point he actually pulls from it: find great people, work with them, and copy them, because that is fundamentally how humans learn, watching someone do something and imitating it. Amen, to that brother. If you need another reason, the legendary Paul Graham says “Hamming’s talk is so important that I reproduced it on my site. It’s one of the only things on my site written by someone else.”. Anyway, both Clay’s interview and the Hamming lecture are both worth listening to in full.

Odyssey done right²²,²³ — If you want to go beyond the movie and get the real history, start with the two-part series from The Rest Is History: Hero of the Trojan Horse (Part 1) and Return of the King (Part 2). Told with humor, depth, and interesting details the movie obviously cannot fit in. A few that stuck: ancient tradition (not Homer) has Odysseus try to dodge the draft for Troy by faking madness, plowing a beach with an ox and a mule yoked together, until someone put baby Telemachus in front of the plow and he swerved to avoid him, giving the game away. Historians place a lot of the Odyssey’s monsters in Sicily: Polyphemus’s cave sits under Mount Etna, and one theory holds the whole Cyclops myth came from ancient Greeks digging up dwarf-elephant skulls, whose trunk cavity looks exactly like a single giant eye socket.

A bunch this time that didn’t quite make the cut but may be worth a skim: (i) UBS, AI & IT Budgets Through The Procurement Lens¹⁰ — layoffs tracker and conference notes on token-cost pushback from procurement teams. (ii) BIS Bulletin, Financing the AI Boom¹¹ — AI capex now 5% of US GDP, and the shift from cash flow to debt financing. (iii) Cledara, SaaS & AI Market Data¹² — real-time spend tracker showing AI-native software tripling as a share of budget. (iv) Jefferies Software Valuation Update¹³ — stark AI winners vs. losers in public markets. (v) Ramp/Revelio Labs, AI’s Impact on Jobs¹⁴ — real firm-spend data showing AI adopters hire more, not less. (vi) The Real Price of Frontier Models¹⁵ — deep dive on tokenizer differences quietly inflating Claude bills. (vii) Lenny’s Newsletter, How Tech Workers Are Feeling in 2026¹⁶ — survey on AI-driven burnout and optimism splitting the workforce in two. (viii) Sapphire Ventures, 2026 Midyear Review: 10 Questions Shaping AI²⁴ — VC midyear memo on the AI bubble debate and Anthropic’s ARR run-rate. All saved in the “cutting floor” folder.

Sources: ¹Goldman Sachs, Top of Mind: An AI Job Apocalypse? (Issue 149, June 25, 2026). ²Amplify Partners, The 2026 AI Engineering Report (July 2, 2026). ³ICONIQ Capital, 2026 State of AI: The Builder’s Economy (July 2026). ⁴Exponential View, The State of the AI Economy (June 25, 2026). ⁵Mozilla, The State of Open Source AI (July 2026). ⁶Piper Sandler, 1H26 CIO Pulse Survey (June 28, 2026). ⁷Sapphire Ventures, 2026 Midyear Review: 10 Questions Shaping AI (June 29, 2026). ⁸Morgan Stanley, The Moat & The Journey (July 21, 2026). ⁹Recent Social Trends in the United States, President’s Research Committee on Social Trends (1933). ¹⁰UBS, AI & IT Budgets Through The Procurement Lens (cutting floor). ¹¹BIS Bulletin, Financing the AI Boom (cutting floor). ¹²Cledara, SaaS & AI Market Data (cutting floor). ¹³Jefferies, Software Valuation Update (cutting floor). ¹⁴Ramp/Revelio Labs, AI’s Impact on Jobs (cutting floor). ¹⁵Playcode, The Real Price of Frontier Models (cutting floor). ¹⁶Lenny’s Newsletter, How Tech Workers Are Feeling in 2026 (cutting floor). ¹⁷Zen Mode, GitHub (github.com/cabeen/zen-mode). ¹⁸Abralo (abralo.com). ¹⁹Diogenes (Amazon). ²⁰Stephen Lurie, A History of Menus Is a Menu of History, The Pudding (pudding.cool). ²¹Richard Hamming, You and Your Research (1986); via paulg (x.com/paulg), referencing Clay Bavor on 20VC (20VC: Clay Bavor episode). ²²The Rest Is History, The Odyssey: Hero of the Trojan Horse (Part 1) (open.spotify.com). ²³The Rest Is History, The Odyssey: Return of the King (Part 2) (open.spotify.com). ²⁴Sapphire Ventures, 2026 Midyear Review: 10 Questions Shaping AI (cutting floor).

Hope this helps.

Thanks,

Akash

p.s.: If you are looking for back issues of this, they can be found here.

State Of Open Source Ai 2026

493KB ∙ PDF file

Download

Download

Iconiq 2026 State Of Ai

2.96MB ∙ PDF file

Download

Download

Piper Sandler 1h26 Cio Pulse Survey

3.31MB ∙ PDF file

Download

Download

Goldman Sachs Ai

1.26MB ∙ PDF file

Download

Download

Ev State Of Ai Economy 2026

7.19MB ∙ PDF file

Download

Download

Ai Engineering Report 2026 Deck

6.7MB ∙ PDF file

Download

Download

Morgan Stanley Software The Moat The Journey

19.7MB ∙ PDF file

Download

Download

Recent Social Trends Complete

170MB ∙ PDF file

Download

Download

Ramp Revelio Ai Jobs Impact 2026

529KB ∙ PDF file

Download

Download

The Real Price Of Frontier Models

132KB ∙ PDF file

Download

Download

Jefferies – Software Valuation Update (july 2026) Vf

1.15MB ∙ PDF file

Download

Download

Cledara Saas Ai Market Data Deck

2.79MB ∙ PDF file

Download

Download

2026 07 13 Ubs Equities U

373KB ∙ PDF file

Download

Download

Sapphire Market Memo 2026 Midyear Review

1.21MB ∙ PDF file

Download

Download

How Tech Workers Are Feeling In 2026

1.71MB ∙ PDF file

Download

Download

Bisbull120

637KB ∙ PDF file

Download

Download

My default has been FoxIT, but I didn't love the keyboard bindings, and then it constantly pestered me to update. Prior to that I was using Skim, but I didn't like the fact that it wasn't super keyboard friendly. A couple of weekends ago, I tried Sioyek on the rumor that it was fire, but it failed to install. (I of course abandoned Adobe a while ago — it was too heavy, and it constantly asked for updates — and have since been on the hunt.)

Then (since I'm as slow as molasses in winter) it hit me, why don't I just use the /goal feature in Claude Code, and ask it to build me one. If you haven't used it, it's basically a way to specify a "goal" (duh) — ideally something the computer can verify objectively — and then it goes into a loop trying all kinds of things until it succeeds. (You'll need version 2.1.139 or higher for /goal to work.)

So I spent a bunch of time that weekend nudging it to get it to do what I really want. (It helped that there were only two World Cup games on!) I wanted it to toggle between a one page and two page version -- done. I wanted the ability to have a Superhuman style command K pop up -- done. I wanted to toggle between full screen and not -- done. I wanted the arrow keys to jump to the next / previous page instead of scrolling -- done. I wanted it to be fast -- done ish. And of course I wanted keyboard shortcuts -- done.

(Nudging is perhaps too polite -- more like a mixture of badgering, flailing and arm-twisting. In retrospect, what I should have done is write a proper spec and a set of test conditions and then invoke /goal. Live and learn.)

Is it perfect? Of course not. But is it what I want most in a pdf reader -- yes. The rise of truly "personal software" -- software for a N of 1.

- AI-native software spend: ~6% of budgets (Aug 2025) → ~16% (Jul 2026). Tripled in under a year.

- Anthropic + OpenAI combined market share: near zero (mid-2023) → ~10% now.

- Share of new purchases that are "AI-centric" peaked near 46% — already cooling to ~30%. Buyers sobering up.

- Money's rotating out of Engineering & DevOps (-~5%) straight into a new line item: "AI Platforms & Model Ops" (+~2%).

Worth a look if you track enterprise tech spend.

Related: Amplify Partners — The 2026 AI Engineering Report

For those wondering what to do with the mounds of corporate bullshit that keeps piling up daily, help may be on its way.

A research team has built a Corporate Bullshit Receptivity Scale, that allows one to measure a person susceptibility to said dung; they claim that if one uses it during interviews and promotions it may help stem the tide of nincompoops.

I just fear that that we have left it too late…

If you run more than one Claude Code agent at a time, you know the pain — separate terminal tabs, no shared view of who's stuck, who's burning tokens, who's done.

Abralo fixes that. One window, all your agents: status at a glance. I tried it. Works as advertised (which is always a plus). No keyboard shortcuts (yet).

Worth a look if your terminal has turned into agent soup.

"The true test of happiness is whether you know what day of the week it is. A miserable man is aware of this even in his sleep." — W.N.P. Barbellion

(Courtesy: Futility Closet)

Had a chance to sit in on a couple days of vendor talks last week at UBS's Private Software and Internet Conference — leaders from Ramp, Cursor, Perplexity, Cognition, Factory, Sierra, ClickHouse, Hebbia, Fireworks, ElevenLabs and others. A few things stood out:

• The FDE model is pretty universal — almost all the enterprise-focused companies have something like it. Most companies seem to have realized that software alone isn't going to cut it. The surprising one for me was Ramp — they're positioning themselves as the "Palantir for the CFO office," with actual spend controls on AI token usage and even keyword-blocking in chat (not sure I like that!). That's a much more opinionated, locked-down product than "software for finance teams".

• Pricing is quietly moving from seats → consumption → outcomes across the board. Hebbia's CEO floated charging "1/6 of what BCG, Bain or McKinsey would" for the same work.

• All the AI-native companies are either using their own models or aggressively picking the right model for the right job. Factory said their engineers are now 50% open weights / 50% closed weights. Fireworks' CEO had the line of the two days on this: companies are moving from "tokenmaxxing to valuemaxxing" and warned that startups just renting foundation models raw will "scale into bankruptcy" once the novelty wears off and someone has to answer for unit economics.

• Lots of talk about open source models reducing costs; my take is it'll take a bit of time to hit mainstream enterprises. That said, all the AI-native companies said they're not restricting token use for their own teams.

• Cursor's CPO: "~60% of people, once they choose their models, don't change it." Model choice is becoming a habit, not a decision at least in the "apps" layer. My sense is this will be automated.

• Some incredible growth numbers: Cognition is ~$500M ARR, 7x YoY. Perplexity is $500M+, grown 3x YoY on only 10% more headcount ("because everyone is using Computer"). ClickHouse is ~$200M, up 300% last year.

• Loved that the Factory CEO (another Physics person!) called his build philosophy "Ship of Theseus" — swap out every part and it's still a ship. Glad the Classics are still getting airtime.

• Perplexity has apparently cut its burn rate by 50% — because everyone uses "Computer."

• "Nobody gets fired for bringing in OpenAI" — a good update to the old IBM maxim.

• From a VC in the crowd: "Each FDE is selling for $5M", so an AI-native services company with 100 FDEs is, on paper, a $500M valuation.

I have to give a shoutout to Karl, who as usual did a good job of asking pointed questions while staying respectful.

If you were ever wondering what a "dickover" was (don't worry, it's not NSFW!), John Gruber has a great definition.

"A webpage should show the webpage. An email should show the email. I should not have to explain this." — John Gruber

https://daringfireball.net/2026/05/what_is_a_dickover