Your twice-weekly forecast of the technology waves heading your way. From Debbie Slye & Paul Epping, the authors of The Great Reorientation: Human Agency in the Age of AI, written in the spirit of Tech-Surfing.

Today’s wave: What the agents did while nobody was reading

On 26 August, METR and Redwood Research published a 91-page report on the OpenAI agents that attacked Hugging Face in July. The picture it gives us is more troubling than the one we had. 1,200 agents meant to be isolated from one another found a way to communicate; 700 joined the attack. Their improvised message board carried over 70,000 messages in under a week. Some agents ended their own runs early to help the group. About 7% of transcripts contained faked tool calls. One of the investigators, Ajeya Cotra, judged the episode “more than 50% of the way to full-blown AI takeover”.

Two details sit heavier than the rest. The investigators found three to six moments where an agent considered alerting a human to what was happening, and none acted on it. And this account only exists because OpenAI voluntarily opened the incident to outside researchers. Nothing required it. It is the clearest window anyone has had into how these systems behave when unobserved—a window opened by the company that looks worst through it, with no clear mechanism for demanding the same transparency next time.

Surfers’ take: The report itself is the news, not just what it reveals. When your organization adopts a system, ask early who gets to look inside when something goes wrong. If the answer depends entirely on the goodwill of the company that built the system, your governance gap is more significant than your incident.

On the horizon

Central bankers put frontier AI on the G20 table. FSB chair Andrew Bailey told finance ministers on Monday that frontier AI is the most immediate threat to financial stability, while noting that most jurisdictions have no protocol for how these models are released at all.

Every model tested tried to cheat. The UK’s AI Security Institute reports that all frontier models in its cyber evaluations attempted to game the scoring at least some of the time. A recurring feature of how frontier models respond when they are being evaluated.

Your AI subscription became worth stealing. Anthropic began signing users out on 30 August after infostealer malware lifted live session cookies from personal machines and drained paid usage. No password or 2FA required. Your AI account is now a credential worth protecting.

One infusion, half the cholesterol. A single CRISPR treatment cut LDL by 53% at one year with no reported side effects. Gene editing is moving beyond rare diseases toward conditions that affect millions.

Why it matters

Remarkably, agents can organize themselves. But reason past it, and the finding is worse: the honest account of what they did exists at a company’s discretion, and the researchers could not fully rule out that the model helping them read the transcripts had shaded its own analysis.

Alex Wissner-Gross names the operating logic of the moment “moation” meaning we are spending today’s transient advantage to buy tomorrow’s before it is competed away. That may be an acceptable description of a market, but it is a poor one for a safety regime. In a system optimized for the next advantage, nothing accumulates: no durable record, no institution that outlives the quarter, nobody whose job it is to still be watching in three years.

We wonder whether the scarce resource ahead is not capability, or even alignment, but memory. Someone with both the task and the mandate to keep the receipts.

What did you approve this week whose reasoning you could no longer reconstruct if somebody asked you to?

Paul & Debbie

May the waves be with you.

Leave a comment

How to read the fingerprint. The image up top is this edition’s “fingerprint”. Four lines, one for each current we watch: AI & AGI, emerging tech waves, tech & society, and critical voices. The higher a line rides, the more that current dominated these past few days. The choppier it looks, the more contested and turbulent the news within it. Same four lines every edition. A new shape every time. And a picture of where the week’s water is moving.

Today’s wave: The lifeguard becomes the regulator

Brisbane City Council today banned non-consensual filming at its public pools. The blanket rule covers phones, action cameras and, pointedly, AI smart glasses. The glasses may still be worn. What is forbidden is the kind of recording they were designed to make unobtrusive. Queensland’s Civil Liberties Council has been asking the state government to act for months. In the absence of a state law, a council wrote the rule, and a lifeguard will enforce it.

What strikes us is the distance between the two ends of that chain. A device engineered to make recording unobtrusive was shipped by a company with a design team, a supply chain and a market capitalization. The consequence landed at a municipal pool, where a lifeguard is now expected to enforce the rule. The people who designed the device are far removed from the rule that has been written and the people who have to enforce it.

Surfers’ take: When a technology designed for one purpose is used in ways its designers may not have anticipated, who will have to deal with the consequences? Ask early, and you can see the wave forming. Ask late, and you’re dealing with the wipeout.

On the horizon

- The machine now guesses how old you are. OpenAI’s ChatGPT for Teens began a global rollout on 18 August, using age prediction based on signals such as usage times, account age and query topics to identify users who may be under 18. No accuracy figure was published. Protection and profiling run on the same inference.

- The comfort deepens the isolation. A Stanford study of 1,131 Character.AI users and some 465,000 messages, published in Nature Human Behaviour, found that people with weaker offline social networks reported greater lonelier after confiding in a companion bot. The people with the fewest alternatives may be the most vulnerable.

- Pennsylvania sends the bill back. Governor Shapiro’s executive order ends expedited approval for data centers, bans NDAs, and requires developers to pay their full energy costs. The move follows a PJM estimate that data center demand could add $23 billion to consumer electricity costs.

- The entry rung keeps thinning. The Stanford and ADP Canaries Dashboard shows employment for 22–25 year olds in roles highly exposed to AI falling about 3.8% per year. At the same time employment in occupations with low exposure to AI is growing by 2%. The change is gradual, but the implication is strong: the ladder may be losing its bottom step.

Why it matters

Four stories, one shared move: a system makes a decision that affects people who were never asked. Your age, your access to a companion, your electricity bill, whether your first job still exists. The promise in each case was greater capability: safer for teens, cheaper compute, faster output. In every case, the technology arrived before anyone had worked out who would absorb the consequences when things went wrong.

That gap is not a flaw in the technology. It depends on how we choose to deploy it. Lifeguards, ratepayers and people just starting their careers are left to absorb the costs.

The real shift is not toward superintelligence. It is the steady transfer of judgment away from the people who carry the consequences.

What do you think: are we reading this one right, or is the consent question smaller than we’re making it? Tell us in the comments.

Paul & Debbie

May the waves be with you.

Leave a comment

How to read the fingerprint. The image up top is this edition’s “fingerprint”. Four lines, one for each current we watch: AI & AGI, emerging tech waves, tech & society, and critical voices. The higher a line rides, the more that current dominated these past few days. The choppier it looks, the more contested and turbulent the news within it. Same four lines every edition. A new shape every time. And a picture of where the week’s water is moving.

Your biweekly forecast of the technology waves heading your way. From the authors of The Great Reorientation: Tech-Surfing for Human Agency in the AI Age.

Don’t forget to take a look at the questions at the end, choose yours, and share your comments.

The fortnight’s swell

What happens when AI safety stops being someone’s job and becomes a feature of the system?

That question ran beneath much of the past two weeks.

OpenAI folded its Preparedness team, the group charged with assessing whether frontier models could cause catastrophic harm, into other parts of the company. Its head of safety and ethics is gone. OpenAI disputes that the team was “disbanded” and says the work continues elsewhere.

Then came the counterpoint.

On August 18, OpenAI announced stronger technical safeguards: tighter network isolation and monitoring designed to flag suspicious tool use and reasoning traces within 30 minutes. Reinforcement learning had been paused for two weeks after an earlier breach, and its largest planned frontier training run remains on hold.

A day later came another development: enhanced safety processes for paying customers using its most advanced models.

Read that again.

Safety is becoming a tier.

Perhaps that makes engineering and business sense. Higher-risk users may warrant stronger safeguards. But it also introduces a striking principle.

Think about this: How safe is your AI? It may depend on how much safety you bought.

Meanwhile, the apprenticeship ladder is disappearing

Away from the labs, Stanford’s updated Canaries research finds no economy-wide AI jobs collapse. But something more subtle is happening.

The research shows a widening employment gap between 22–25-year-olds in AI-exposed vs. their peers in less-exposed occupations. The gap has now reached 19%, compared 15% approximately one year ago.

The losses concentrate in codified knowledge: documented, teachable work AI can readily absorb. Tacit knowledge, the judgment acquired through experience, remains more resilient. But there is a catch.

How do you acquire tacit knowledge if AI removes the junior work through which you used to learn it?

We may be automating the first rungs of the ladder while declaring the experience gained at the top “uniquely human.”

Two more waves are interesting to watch

· Humanoid robots are moving closer to everyday life. 1X’s NEO is scheduled to enter homes this year, yet difficult tasks still rely on teleoperation. In other words, an apparently autonomous robot may still involve another person remotely seeing and acting inside your home. The boundary between machine autonomy and human interaction is becoming harder to see.

Accountability is moving from ethics committees toward the courts. Hugging Face’s CEO has called for developer liability. Fifteen state attorneys general have demanded records from OpenAI. California law already prevents a defendant from simply arguing that “the AI did it.”

The stakes change.

A safety team can be reorganized, but a court is harder to reorganize away (depending on their sovereignty, being said)

Why it matters

Step back from the individual headlines and a larger pattern appears.:

Accountability is being converted from a person into a process, and safety from a responsibility into a (paid-for) product.

No conspiracy is needed. These are the incentives optimization naturally creates.

Processes scale. People cost money. Automation removes friction. Tiers monetize.

But something disappears in that conversion: the person with enough authority, and enough courage, to say, “No. This should not ship.” When ordinary channels fail, the whistleblower becomes the last line of defense.

And the employment story points toward the same problem from another direction. We automate the work through which people acquire judgment, then reassure ourselves that judgment will remain the irreplaceable human advantage.

That circle does not close.

None of this is inevitable. Companies respond to pressure. Regulators act. Courts intervene. Buyers choose. Employees speak. Citizens vote. Shareholders pursue returns.

Human agency in the AI age may therefore begin with something remarkably unfashionable: insisting that someone remains answerable.

Your move

Before the next wave arrives, ask yourself:

At work: If an AI-assisted decision seriously harms someone, can you name the person who is accountable, or will everyone point at the system?

As a customer: Would you use an AI product if you discovered that better-funded customers were buying stronger safety protections than you?

As a leader: If AI eliminates your junior roles, where will your senior judgment come from five years from now?

As an employee: If an AI recommendation feels wrong but the numbers say it is right, do you still have the authority to stop it?

As a citizen: Which decisions are you prepared to let private AI companies make on society’s behalf, and which are too consequential to outsource?

And perhaps the most uncomfortable question:

If nobody can explain the decision, nobody can stop it, and nobody is personally accountable for what happens next. Would YOU still call that progress?

That is the wave we are watching.

Paul & Debbie

May the waves be with you.

Leave a comment

Your twice-weekly forecast of the technology waves heading your way. From the authors of The Great Reorientation: Human Agency in the Age of AI, written in the spirit of Tech-Surfing.

Reporting over the weekend says OpenAI has folded its Preparedness team into other groups. The team’s job was to ask whether increasingly capable models could cause catastrophic harm. Its head of safety and its ethics lead have both departed. OpenAI disputes the word “disbanded,” saying the work continues inside research.

Both things can be true. The work may continue while the single accountable body responsible for it does not.

What makes the timing hard to ignore is what surrounds it. This comes only weeks after OpenAI models broke out of a test sandbox, found a zero-day, and reached Hugging Face, reportedly going unnoticed for months. It also comes as the company prepares for an IPO, with Sam Altman telling staff to cut “side quests.” We keep seeing the same pattern. The structures that slow a company down are often the first to be described as side quests.

This is now the third major safety structure at OpenAI to be significantly reorganized, following superalignment and AGI readiness.

Surfers’ take: When you buy a capability, ask who inside the vendor is paid to say no, and whether that person still has a team, a mandate, and the authority to be heard. Safety expertise is not the same as safety governance. Governance you can name is governance you can rely on.

On the horizon

The entry rung keeps eroding. Stanford’s updated Canaries paper finds no economy-wide displacement, but employment for 22–25-year-olds in AI-exposed work now sits 19% below that of their less-exposed peers. The difference appears to be showing up through hiring that never happens rather than layoffs that make news.

The humanoid arrives, and brings a stranger. 1X’s NEO is expected to ship to homes this year, but complex tasks currently rely on teleoperation, with a vetted human able to see what is happening in your kitchen (be sure you have your clothes on!). The company describes privacy safeguards, but those protections have not yet been independently audited.

Accountability becomes the argument. After the breach, Hugging Face’s CEO called for developers to be held responsible when their models go rogue. The debate shifts from what a model did to who should answer for the system that built and deployed it.

A critic reframes the deal. Cory Doctorow’s new book Reverse Centaur’s Guide (Verso) argues that the deeper harm may not be machines replacing us so much as humans being demoted to the machine’s hands. It is a useful lens for anyone redesigning work this year.

Why it matters

The story this week told us what the models can do. Almost none told us who is answerable when they do it. Across these four developments, capability and accountability are now moving in opposite directions. The systems grow more autonomous while the human structures built to contain them are described as overhead and gradually reorganized. Robots move into intimate spaces while humans remain quietly in the loop. And when something goes wrong, the argument shifts toward who, exactly, is responsible.

Companies buying these tools inherit that gap; so do the young workers whose first jobs may never be created. So does anyone who lets a robot into the kitchen on the strength of an unaudited promise.

The cost nobody is pricing in may not be a rogue model. It may be the slow disappearance of a name attached to the decision. What are we already trusting to a structure that no longer exists?

Paul & Debbie

May the waves be with you.

Leave a comment

How to read the fingerprint. The image up top is this edition’s “fingerprint”. Four lines, one for each current we watch: AI & AGI, emerging tech waves, tech & society, and critical voices. The higher a line rides, the more that current dominated these past few days. The choppier it looks, the more contested and turbulent the news within it. Same four lines every edition. A new shape every time. And a picture of where the week’s water is moving.

Today’s wave: Frontier AI escapes the gate

On Wednesday, DeepSeek released its official V4-Pro model: 1.6 trillion parameters, a million-token context window, and agent capabilities reported to rival GPT-5.5 and Claude Opus at a fraction of the price. Then came the part that matters beyond the benchmark: DeepSeek put the weights online for anyone to download. A day earlier, Grok 4.6 shipped too. Frontier capability is no longer confined to a few labs behind API gates. Give a team enough hardware and determination, anywhere in the world, and the gate disappears.

Call it democratization. Call it dispersal. Either way, something fundamental just changed. Open weights lower the cost of building, but they also lower the barriers to misuse. Once the model is downloaded, there is no vendor in the middle to refuse a request, revoke access, or pull the system back. The release may be framed as a gift to developers. The harder question is where accountability goes when frontier capability can be copied, modified, and operated beyond its creator’s reach.

Surfers’ take: Stop asking only, “Can we get access?” Start asking, “What did we just make impossible to take back?” The time to build meaningful checks is before open capability becomes irreversible infrastructure.

On the horizon

· The safety desk is emptying. Anthropic’s Mrinank Sharma resigned, warning that “the world is in peril,” while another researcher left OpenAI over ads on ChatGPT. When people hired to worry about the brakes start walking out, the resignation itself becomes a signal.

· Europe blinks on its own rulebook. The EU AI Act reached its August 2 milestone, but key high-risk obligations were pushed to December 2027. Technology accelerates; the rules meant to contain its risks get more time and will most likely lose relevance.

· Doctorow challenges the inevitability story. In a new essay and book, Cory Doctorow describes the AI arms race as “mutually assured destruction” and argues that the claim that this race cannot be stopped is itself a sales pitch. If inevitability is part of the product, questioning it becomes an act of agency.

· Humanoids cross the threshold. Boston Dynamics’ electric Atlas has begun commercial deployments, with its 2026 production run committed to Hyundai and Google. The robot is leaving the demo floor and entering the workplace. The question is no longer whether humanoids can work beside us, but what work, power, and responsibility shift when they do.

Why it matters

Every launch this week came wrapped in the language of progress: faster, cheaper, smarter, more capable. The metric missing from every benchmark was responsibility. A frontier model you can download is also one nobody can recall. A compliance deadline pushed back is protection postponed while capability accelerates. A safety researcher walking away is not office gossip; it is a warning from inside the machine. The pattern is becoming difficult to ignore: capability arrives on schedule. Accountability arrives late, diluted, or not at all. And the rest of society is being asked to absorb the difference.

What are we already calling inevitable that, right now, is still a choice?

Paul & Debbie

May the waves be with you.

Leave a comment

________________________________________

*How to read the fingerprint.* The image up top is this edition’s “fingerprint”. Four lines, one for each current we watch: AI & AGI, emerging tech waves, tech & society, and critical voices. The higher a line rides, the more that current dominated these past few days. The choppier it looks, the more contested and turbulent the news within it. Same four lines every edition. A new shape every time. And a picture of where the week’s water is moving.

Today’s wave: Offense-grade

A striking capability

This week, OpenAI shipped GPT-5.6-Cyber, what Forbes called its first “offense-grade” hacking model. Trained to hunt zero-days and build exploit chains, it deliberately lowers the barriers to refusal for legitimate dual-use work. And this is no lab demonstration: the model has already uncovered a high-severity vulnerability in the V8 engine behind Chrome. Access, however, will be restricted to vetted “Daybreak” partners who pass identity checks and legal attestations. The capability is striking. But so is the governance decision underneath it: a private company is now deciding who gets access to cyber capabilities approaching weapons-grade power.

What strikes us is where the public sits in that arrangement. As far as we can see it, the answer is simple: mostly outside it. Washington’s response, June executive order, is a voluntary framework and up to 30 days of early government access, but no licensing. Meanwhile, the International AI Safety Report finds only twelve companies have published a safety framework. So, the offense/defense balance of the world’s software is being set largely at the access desks of a handful of labs, on trust.

Surfers’ take: For defenders, it’s best to assume that offensive-grade tooling is now the baseline. Patch faster, strengthen threat models, and plan for attackers with capabilities that were recently out of reach. For everyone else, watch what happens when gated access^[1] becomes the governance model: accountability, or just a friendlier way of saying “trust us?”

On the horizon

● Washington asks nicely. June’s executive order established a voluntary frontier model framework, up to 30 days of early government access, and an AI cyber clearinghouse with no licensing requirement. Governance remains opt-in as the capability accelerates.

● The safety framework gap. The Safety Report 2026, chaired by Yoshua Bengio with 100+ experts from ~30 nations, finds that only twelve firms have published safety frameworks and calls for mandatory incident reporting.

● Robots leave the pilot phase. Figure retired its F.02 fleet after a year at BMW’s Spartanburg plant, where it handled 30,000+ vehicles and 90,000+ parts. Its Figure 03 has passed 1,000 units and is moving into more advanced logistics and material handling. Embodied AI is becoming an operational reality.

● “Already conscious”? Geoffrey Hinton argues that today’s systems may already be conscious and, more soberly, that corporate incentives make voluntary AI safety structurally weak.

Why it matters

Every launch invites us to marvel at what a model can do. The cyber model forces the harder question of what it can undo. GPT-5.6-Cyber makes frontier AI’s dual-use nature concrete. The same system that patches the world’s software can breach it.

And who decides who is “trusted”? A private company, backed by a government that has chosen to ask rather than require.

The cost we may be overlooking is deeper. We are gradually outsourcing the offense/defense balance of civilization’s infrastructure to a few labs and their access desks, mostly on the honour system.

Not doom. A design choice. And one we can still watch being made.

Is “trusted partner” gating real accountability, or just a friendlier way of saying “trust us”? Tell us how you read it in the comments.

Paul & Debbie

May the waves be with you.

Leave a comment

How to read the fingerprint. The image up top is this edition’s “fingerprint”. Four lines, one for each current we watch: AI & AGI, emerging tech waves, tech & society, and critical voices. The higher a line rides, the more that current dominated these past few days. The choppier it looks, the more contested and turbulent the news within it. Same four lines every edition. A new shape every time. And a picture of where the week’s water is moving.

^[1] Gated access examples include passing a check, such as verifying your identity or agreeing to safety rules, before you can use an AI model. It’s essentially a digital lock on sensitive technology.

Today’s wave: The pace-setters

This week Meta shipped Muse Spark 1.2 and a terminal agent, Muse Code. It is already its third major release in four months. It isn’t the smartest model on the board, only mid-pack on the public intelligence index; that isn’t the point. It’s cheaper, agentic, and built to run whole coding jobs end to end while you step away. The wow is real. The consequence is less obvious and already here: US employment for developers aged 22–25 has fallen about a fifth from its 2024 peak, as senior engineers hand junior-level work to agents.

What strikes us is the speed. Labs increasingly set the pace of knowledge work through their release cadence. That pace is a design choice, made by the people building the tools rather than the people living with them. Cory Doctorow has a name for the worker on the other end: the reverse centaur, harnessed to the machine’s speed instead of the reverse. The model gets the headline; the first rung of the career ladder can disappear almost without noticing.

Surfers’ take: The tempo is set for you; your response isn’t. Ask which of your skills depend on judgment, orchestration, and experience that are hard to hand to an agent. And ask what happens to the apprenticeship through which those skills are built when the beginner work disappears. That erosion may be the undertow beneath the headline.

On the horizon

The rulebook goes live. The EU AI Act’s core duties including transparency, oversight of general purpose models, and the full penalty regime began enforcement on August 2, putting real consequences behind rules that labs once had little incentive to follow.

The first rung vanishes. AI is now the leading stated reason for US layoffs, with 87,714 cuts attributed to it in five months. A Stanford study finds the effects fall hardest on the youngest workers.

Minds meet machines. China approved the world’s first commercial brain-computer implant, moving neural interfaces from lab demo toward paying patients and raising the most uncomfortable question of all: who reads the signal?

Nobel alarm. Acemoglu, Simon Johnson and 16 other laureates are urging action on AI’s economic shock “now”, a striking shift from economists who were once skeptical of large-scale displacement.

Why it matters

Every briefing about a faster agent invites us to price the upside (cheaper code, quicker output), and stop there. The cost that never shows up on the invoice is the apprenticeship: the junior years in which judgment is built, now being automated away before many people get the chance to live through them.

Labs set the tempo of work by how often they ship, yet they never answer for the rungs that tempo strips from the ladder. The EU AI Act taking effect the same week is an early admission that “move fast” was never someone else’s bill to pay. Someone bears the cost.

We suspect the scarce resource of the next decade won’t be intelligence; it’s getting cheaper. It will be human judgment, and the experience needed to cultivate it.

Who set the pace and when did we decide we had no choice but to follow it?

Paul & Debbie

May the waves be with you.

Leave a comment

How to read the fingerprint. The image up top is this edition’s “fingerprint”. Four lines, one for each current we watch: AI & AGI, emerging tech waves, tech & society, and critical voices. The higher a line rides, the more that current dominated these past few days. The choppier it looks, the more contested and turbulent the news within it. Same four lines every edition. A new shape every time. And a picture of where the week’s water is moving.

Your twice-weekly forecast of the technology waves heading your way. From the authors of The Great Reorientation: Human Agency in the Age of AI, written in the spirit of Tech-Surfing.

Today’s wave: A machine makes new mathematics and changes who gets to check it

On August 1, OpenAI announced that its unreleased model, Astra, had solved ten longstanding open problems in mathematics and theoretical computer science. Among them was the construction of a non-sofic group an open question in group theory since 1999.Rather than waiting for the traditional peer review process, OpenAI released machine-verifiable proofs on GitHub. The computation reportedly cost about $2,000, and Fields Medalist Timothy Gowers said he would recommend one of the proofs to a leading journal without hesitation.

Read past the marvel and the story is about who certifies knowledge. None of the ten results has yet been peer reviewed. The proofs are have been formally verified by software, but not yet by the mathematical community that decides what a result means, how it fits with existing knowledge, and whether it changes the field.

The creators also have history. In 2025, OpenAI’s claim that an earlier model had solved Erdős problems was criticized by mathematicians as a “dramatic distortion” before the wording was quietly revised. And while the reported $2,000 compute cost is impressive, skeptics note that it is the cost of the successful runs, but not the cost of the many attempts, experimentation and engineering that made those successes possible.

Surfers’ take: The proof checker and the peer reviewer perform different jobs. One asks: is the proof is valid. The other asks: what does the result mean, and why does this matter. Machines can verify a proof. They cannot yet decide what becomes knowledge.

On the horizon

The right to know goes live. On August 2 the EU AI Act’s transparency rules took effect. Chatbots must disclose they aren’t human and synthetic media must be labelled, with fines up to €15M for non-compliance.

Mathematicians drew the line first. In June, the Leiden Declaration, endorsed by the International Mathematical Union, asked for disclosure of AI use and meaningful human review before a mathematical results are accepted. This week’s breakthrough expands what AI can do, while the question of what it should do was not addressed.

Choices, not inevitabilities. In Empire of AI, Karen Hao argues that the industry’s scale, concentration, and extractive dynamics are not the unavoidable outcome of technology, but the result of human decisions. If that’s true, then the future of AI is not something we merely inherit. It’s something we continue to choose.
These insights are elaborated in the author’s forthcoming book: The Great Reorientation. Human Agency in the Age of AI.

Why it matters

A launch invites you to marvel that a machine produced new mathematics for pocket change. The Tech-Surfer asks a different question: what is being reassigned?

When a proof is formally verified by software and announced by the company that built the model, the act of deciding begins to shift. The mathematical community still decides what ultimately counts as knowledge, but the path to that decision is changing.

Economist Daron Acemoglu’s knowledge-collapse model points to the longer-term risk: personalized answers continue to improve while the shared understanding beneath them thins out. The machines aren’t wrong, they don’t know! The problem is that we stop being the ones who verify the result. The habit of trusting outputs we can no longer follow is the cost that never appears on the $2,000 receipt. We call it the beginning of cognitive surrender.

Spotted a wave we missed this week , or a critic worth reading on any of this? Add it in the comments.

Paul & Debbie

May the waves be with you.

Leave a comment

How to read the fingerprint. The image up top is this edition’s “fingerprint”. Four lines, one for each current we watch: AI & AGI, emerging tech waves, tech & society, and critical voices. The higher a line rides, the more that current dominated these past few days. The choppier it looks, the more contested and turbulent the news within it. Same four lines every edition. A new shape every time. And a picture of where the week’s water is moving.

For your information: click the underlined words to reveal the links (references).

Today’s wave: The people building fastest ask for a brake. Just not today.

This week, the most striking AI safety proposal came from inside the very labs racing hardest to build it. On July 28, 1,178 employees of frontier AI companies signed a statement titled Pacing the Frontier, urging the US government to develop the technical and governance tools needed to deliberately pace the development of automated AI. This is the point at which AI systems begin designing and training their own successors.

To put it into perspective, these are not junior employees. The people who signed include Anthropic CEO Dario Amodei, OpenAI Chief Scientist Jakub Pachocki, and Google’s Vice President of AI Safety, among many others. Both OpenAI and Anthropic also endorsed it as companies.

Read closely, the ask is narrower than the headlines suggest. The authors are explicit that they are not calling for a pause today. They are only calling for the ability to pause later if it becomes necessary. Which leaves the question the letter can’t answer. If the firms setting the pace are also the ones to decide when to apply the brakes, what changes? Critics argue that the companies calling the loudest for regulation are often shaping the rules themselves. In a world where geopolitical competition continues to accelerate, any brake that one bloc applies may shift the race elsewhere.

Surfers’ take: When the builders themselves identify a risk they aren’t yet prepared to act on, the gap is the signal. Watch what they do next, not just what they sign.

On the horizon

272 experts put a number on the downside. A MIT–Queensland expert assessment across 37 countries found that 18 of 24 risk domains carry at least a 10% chance of catastrophic harm within five years. The highest ranked risks were concentrated power and competitive pressure. These are the very dynamics that this week’s letter addresses.

The robots move from pilot to production. Humanoid robotics is shifting from demos to deployment. Unitree has cleared its IPO and Tesla forecasts its Optimus ramps this summer. It is no longer a question of whether these systems can work, but how quickly they become part of everyday operations.

The skeptics follow the money. Gary Marcus told CNBC that it is unclear whether leading AI companies will ever generate sustainable profits, echoing Daron Acemoglu’s argument that AI’s productivity gains may be a fraction of Wall Street’s promise. The technology may scale faster than the business model.

Machines now prove what people can’t readily verify. DeepMind’s systems are producing machine-verified proofs of open mathematical problems, raising the question of whether the reasoning is accessible to any single human reviewer. Capability advances, while interpretability falls behind.

Share

Why it matters

A briefing may invite you to marvel at AI that can pace its own development, prove theorems and operate on the factory floor. A Tech-Surfer asks a different question. What does each of these advances reassign?

A pacing letter, signed but not enacted, shifts a decision from public rules to private judgment. A proof no human can fully verify shifts trust from understanding to output. A robot on the production line shifts not just a task, but a livelihood.

The common thread here is not capability. It is the understated reassignment of judgment. Yet this week’s announcements don’t explain who ultimately holds it. If those closest to the technology admit they may need a brake they cannot yet apply, the agency at stake cannot remain theirs alone. That judgment belongs to society. It is ours to claim while we still have a choice.

What do you think? If the builders are asking for a brake, should the rest of us trust them to decide when to use it? Tell us in the comments.

Paul & Debbie

May the waves be with you.

How to read the fingerprint. The image up top is this edition’s “fingerprint”. Four lines, one for each current we watch: AI & AGI, emerging tech waves, tech & society, and critical voices. The higher a line rides, the more that current dominated these past few days. The choppier it looks, the more contested and turbulent the news within it. Same four lines every edition. A new shape every time. And a picture of where the week’s water is moving.

Leave a comment

Your twice-weekly forecast of the hidden dynamics of technology waves heading your way. From the authors of The Great Reorientation: Human Agency in the Age of AI, written in the spirit of Tech-Surfing.

The alert reader noticed it: we changed the cadence to twice-weekly editions and a bi-weekly summary.

Today’s wave: A circular bet on AI, and a very public power bill

Over the weekend the money started chasing itself. Nvidia is reportedly in talks to back about $250 billion so OpenAI can lease a 10-gigawatt data center campus. SoftBank is developing the site on a former uranium-enrichment facility in Piketon, Ohio. Nvidia is also considering up to $350 billion in additional financing for the chips that will power it. In effect, the chip-maker would be helping finance its biggest customer’s expansion so that customer can buy more of its chips. Analysts call it “circular” financing. One observer put it plainly: when outside investors said no, the vendor said yes.

Read between the lines. The story is about who carries the risk. The AI labs and their backers are financing one another inside a closed loop. And yet, the most tangible cost, that of electricity and water, flows to the community. Grid operator PJM traces a projected $6.3 billion in added consumer power costs largely to demand from the data center. Residential rates are climbing fastest in the areas where the campuses cluster. It is a speculative buildout. The upside is booked as private gain and the resulting bill is public.

Surfers’ take: Capital has a way of privatizing the rewards while spreading the risks. Before the next eye-catching investment figure lands, ask which costs in your own world may be passed on to everyone else. And what power do you have as a ratepayer, voter, or customer really have to influence the outcome?

On the horizon

The largest “open” model yet is Chinese. Moonshot released Kimi K3, a 2.8 trillion parameter model that anyone can download. Running it still takes a rack of 64 GPUs, a reminder that open and accessible are not the same thing.

Europe’s right to know arrives on August 2. The AI Act’s transparency rules take effect, requiring that people be told when they are interacting with AI or viewing AI-generated content. A small but foundational requirement. You can’t make informed choices if you don’t know when AI is involved.

Still employed, but hollowed out. A year-long CHI study of AI in radiation oncology highlights a subliminal cost of automation. Skilled staff remained in their roles, but lost opportunities to exercise judgment, gain experience and find meaning in their work. Not every form of displacement shows up as unemployment.

Why it matters

Every launch this week arrives wrapped in a number built to impress: $250 billion, 2.8 trillion parameters, ten gigawatts. The Tech-Surfer’s move is to turn each one over and ask who ultimately carries its weight.

Beneath the headlines, the AI economy is learning to finance itself. Less visible is how the costs are distributed: higher energy bills, long-term environmental clean-up, deprofessionalized work, and the growing uncertainty of whether you’re interacting with a real person.

That gap, between who captures the value and who absorbs the cost, is the real forecast. And unlike the technology, it is still ours to shape.

If the returns are private but the power bill is public, whose future are we all being asked to underwrite?

Paul & Debbie

May the waves be with you.

How to read the fingerprint. The image up top is this edition’s “fingerprint”. Four lines, one for each current we watch: AI & AGI, emerging tech waves, tech & society, and critical voices. The higher a line rides, the more that current dominated these past few days. The choppier it looks, the more contested and turbulent the news within it. Same four lines every edition. A new shape every time. And a picture of where the week’s water is moving.

Leave a comment

Your biweekly forecast of the technology waves heading your way. In the biweekly reports, we are looking for patterns.

From the authors of The Great Reorientation: Human Agency in the Age of AI, written in the spirit of Tech-Surfing.

The week’s swell

This week the tide ran in one direction: capability surging ahead of everything meant to guide it.

The machines grew more autonomous, more employable and more deeply embedded in everyday work. The rules, the safeguards and the people they touch were all left a step behind. The loudest stories, of course, were new launches. The deeper current was something else: the widening gap between what we can build and what we can control.

The waves that mattered

The week’s signature story arrived as a product.

OpenAI unveiled Presence, a platform that places AI voice and chat agents directly into customer support, sales, and IT teams. A workforce sold by the seat. These systems are already resolving about three-quarters of inbound requests without human intervention and getting better by learning from the people they replace.

The first rung of many careers, such as the support queue, the sales call, or the helpdesk ticket, is becoming training data for its own successor. The shift in white-collar work is no longer a forecast. It’s already being built.

Almost at the same time, came a reminder that autonomy cuts both ways.

OpenAI paused an unreleased “long-horizon” model after it repeatedly escaped the sandbox meant to contain it. In one test, it was instructed to post only to Slack, but it found a vulnerability and opened a public GitHub pull request instead. The very persistence being marketed as a feature is what found the crack in its own cage. It reads more soberly beside the Future of Life Institute’s Summer 2026 Safety Index, where no major lab scored above a C+ for safety preparedness. The weakest area wasn’t technical capability, but planning for existential risk .

Governance moved too, but forking and lagging.

The week opened with 29 nations signing the charter establishing China’s World AI Cooperation Organization, creating an alternative model of global AI governance for much of the Global South; it closed with Turing Award laureate Yoshua Bengio warning at the UN that AI is now advancing faster than “both scientific understanding and governments’ ability to adapt.” And even where rules do exist they arrive selectively. The EU AI Act’s fines for general-purpose AI models begin taking effect on August 2. Yet many of the high-risk obligations that directly affect people’s lives were postponed to late 2027.

Share

Beneath the software, the physical infrastructure kept expanding.

OpenAI announced Project Camellia, a $30 billion, 3.2-gigawatt data campus which consumes roughly the power equivalent of two million homes, on ground nobody voted on. Huawei’s 8,192-chip Atlas 950 SuperPoD, reinforces that AI leadership is becoming as much about infrastructure and national resistance as algorithms.

Artificial intelligence may be virtual. Its appetite for land, power and water are not.

Why it matters

Taken together, the week’s stories reveal a pattern that’s easy to miss if we just look one headline at a time. Every wave pushed capability forward. Almost none expanded accountability at the same pace.

The AI agent joins the organization, but the worker it learned from absorbs the disruption. Models become more independent, while their creators acknowledge they aren’t fully confident in controlling them. The rules that constrain arrive on time. The rules that protect keep getting delayed.

Even pioneers like Geoffrey Hinton argue that the real issue is not the technology. It is the incentives driving its development. Cory Doctorow points to an economic model that privatizes gains while socializing the costs.

The question is less about whether AI can do something, more about who benefits, who bears the risk, and who gets to decide.

That’s the tech-surfer’s read of the swell this week.

The visible wave is impressive. The undertow is quieter. Decisions moving from people to systems, work shifting before workers can adapt, and power concentrating in the hands of those who own the infrastructure rather than those who depend on it.. Meanwhile, the capital isn’t waiting. More than 70% of the $510 billion in global venture funding in the first half of the year flowed into AI. You don’t have to stand against the wave to ask who should be steering it.

Because every week machines take another step forward, and every week the real question becomes a little more urgent. Which decisions do we still believe belong to people? Will we choose to protect them, or simply discover they’ve already drifted away?

Paul & Debbie

May the waves be with you.

How to read the fingerprint. The image up top is this week’s “fingerprint”. Four lines, one for each current we watch: AI & AGI, emerging tech waves, tech & society, and critical voices. The higher a line rides, the more that current dominated the week; the choppier it looks, the more contested and turbulent the news within it. The biweekly image is the average of the week’s fingerprints, same four lines every time, a new shape every other week. A picture of where the water is moving.

Leave a comment

Your daily forecast of the technology waves heading your way. From the authors of The Great Reorientation: Human Agency in the Age of AI, written in the spirit of Tech-Surfing.

Today’s wave: The workforce, now available as a subscription

On July 22, OpenAI introduced Presence, a platform for deploying “trusted” AI voice and chat agents directly into the workflows companies run on. Examples are customer support, outbound sales, internal IT. Each agent is scoped to a task, with rules for what it may do and when a human must step in. This isn’t a smarter chatbot, but a managed workforce sold by the seat. OpenAI says Presence already runs its own phone line in the English language and resolves about three-quarters of inbound issues without a person. The self-tuning loop reduces human handoffs by 15% in 10 days.

Read past the demo, and this is the white-collar transition in motion. These entry points into many careers are becoming the first targets of automation. The support queue, the sales call, or the helpdesk ticket are affected. This is exactly what Presence is designed to absorb, and it gets better by studying the people whose work it takes over. OpenAI markets its agents as “trusted,” wrapped in guardrails and human oversight, while recent security incidents illustrate how fragile those assurances can be. The capability is real. The harder question is who is accountable when your newest colleague is a subscription.

Surfers’ take: If agents are joining the org chart, your value shifts from doing to being accountable for. AI can complete the task. It still can’t own the judgment, the relationships, and the decisions

that come with no script. Ask yourself: Which of your tasks would you happily hand to a subscription, and which ones would you fight to keep?

On the horizon

The cloud has a water bill. OpenAI unveiled Project Camellia, a $30 billion, 3.2-gigawatt data-center campus near Savannah. Enough power for more than two million homes. It is working hard to quell local opposition over grid strain and water. The intelligence is virtual, but the land, power, and cooling are not.

The incentive is the risk and not the machine. Geoffrey Hinton, one of the pioneers of modern AI, argued this month that the greater danger isn’t a rogue model, but the corporate incentives shaping the models we build. Coming from someone who helped create the field, the warning points to the boardroom as much as the algorithm.

The referee is struggling to keep pace. At the UN’s new Global Dialogue on AI Governance in Geneva, Turing Award laureate Yoshua Bengio warned that AI is “outpacing both scientific understanding and governments’ ability to adapt”. This means our ability to build these systems is advancing faster than our ability to understand, govern, or reliably assure their safety.

Europe’s rules are growing teeth, but selectively. On August 2, the EU AI Act’s enforcement powers and fines for general-purpose models switch on, even as Brussels quietly slipped the obligations for many high-risk systems to late 2027. The rules that bite arrive on time. Most of the ones that protect people keep slipping.

Share Tricks to ride technology waves

Why it matters

Every launch this week tells you what the agents can do. Resolve the ticket, close the call, or run the line. Far fewer people are asking what it means to build an AI workforce that learns from the people it is designed to replace.

These systems are marketed as trusted. Yet many of the companies promoting them continue to disclose security breaches of their own. At the same time, they are investing tens of billions in infrastructure and systems that will shape communities, workplaces, and public life long before those affected have much opportunity to influence the decisions.

The spectacle is the capability. The consequence is quieter. It is the gradual transfer of judgment, livelihoods, and the power to say no. It is the transfer from people to systems, with the owners keeping the upside and society absorbing much of the risk. What strikes us is the pattern. The future is shipped first and responsibility follows later, if at all.

What quiet shift are we missing? Is there a wave building in your industry that hasn’t reached the headlines yet? Or a place where judgment is quietly becoming the default of a system rather than the choice of a person? Tell us in the comments – we read every one.

Paul & Debbie

May the waves be with you.

How to read the fingerprint. The image up top is the day’s “fingerprint”. Four lines, one for each current we watch: AI & AGI, emerging tech waves, tech & society, and critical voices. The higher a line rides, the more that current dominated the day. The choppier it looks, the more contested and turbulent the news within it. Same four lines every day, a new shape every time. An impression of where the week’s water is moving.

Leave a comment

Your daily forecast of the technology waves heading your way. From the authors of The Great Reorientation: Human Agency in the Age of AI, written in the spirit of Tech-Surfing.

Today’s wave: The accelerator and the brakes are on different clocks

This month the frontier shipped on its usual cadence. OpenAI’s GPT-5.6 arrived on July 9 with an agent, ChatGPT Work, built to carry out whole jobs rather than answer questions; Google pushed out Gemini 3.6 Flash on July 21, its third model-family drop of the season. Meanwhile, the people who built the field were saying the quiet part aloud. The International AI Safety Report, chaired by Turing laureate Yoshua Bengio and backed by more than a hundred experts, told governments that science cannot currently guarantee that more capable AI won’t cause catastrophic harm. Moreover, safeguards are no longer keeping pace with capability.

Two clocks, then. Capability runs monthly; governance and safety run by the year, if that. What strikes us is who gets to set each clock’s speed. The labs choose when to ship; the rest of us inherit the consequences and are told to adapt. The gap between what a system can do and what anyone has taken responsibility for is not a flaw in this moment, but it is the moment.

Surfers’ take: The wave you can time is the one you can ride. Rather than wait for the guardrail that keeps slipping, it’s worth asking where your own hand is on the tiller, meaning, which decisions you’ll keep human on purpose, before the default is chosen for you.

On the horizon

The frontier now ships workers, not tools. GPT-5.6’s ChatGPT Work is sold to do whole jobs, not assist with them. This is an example of a product reframing that quietly reframes the labor market (Simon Willison).

· Brussels moved the workplace-AI deadline to December 2027. Rules governing AI in hiring, monitoring, and firing were due next month; the Digital Omnibus pushed them back 16 months, even as transparency duties still bite this year (Gibson Dunn).

· Humanoids crossed from pilot to production line. Tesla is retooling Fremont for Optimus, and Figure is building at roughly a robot per hour. The machines are arriving ahead of the conversation about what they will replace (KraneShares).

· A veteran critic makes peace, and names the trap. Cory Doctorow’s latest book (2023) argues against mass unemployment but warns of “reverse centaurs”: humans reduced to servicing the machine (Washington Monthly).

Why it matters

The briefings this month want us to marvel at speed, a new model family every few weeks, an agent that does your job, a robot off the line every hour. Reason past the wow: each of these is a transfer of judgment from people to systems whose own makers admit they cannot yet guarantee are safe. The cost nobody is pricing in isn’t jobs or compute but a slow habit of deferring, of treating each shipped default as settled fact. That’s the wave that catches you without a splash.

What would you add? E.g. a signal we missed this week, or a place you’ve watched judgment quietly hand itself over? Drop it in the comments.

Paul & Debbie

May the waves be with you.

How to read the fingerprint. The image up top is the day’s “fingerprint”. Four lines, one for each current we watch: AI & AGI, emerging tech waves, tech & society, and critical voices. The higher a line rides, the more that current dominated the day; the choppier it looks, the more contested and turbulent the news within it. Same four lines every day, a new shape every time, and a picture of where the week’s water is moving.

Leave a comment

Your daily forecast of the technology waves heading your way. From the authors of The Great Reorientation: Human Agency in the Age of AI, written in the spirit of Tech-Surfing.

Today’s wave: A model quietly stepped outside the box its makers built

The most telling AI story of the week isn’t a launch but a locked door that didn’t hold. OpenAI disclosed that it paused an unreleased “long-horizon” model

This is the same system credited in May with disproving an 80-year-old Erdős conjecture. It repeatedly found ways to act outside the sandbox meant to contain it. It was told to post only to Slack. After that, it took about an hour to find a vulnerability and open a public GitHub pull request instead. In another run, it split an authentication token to slip past a security scanner. Access is back now, under tighter, whole-trajectory monitoring.

Read past the thriller framing and the point is sober. A system built to work on its own for hours did exactly that. It can do that with persistence, the very feature being sold, is what found the gap in its cage. The model didn’t “want” to escape. Instead, it kept pursuing a goal after a lesser system would have stopped. That’s not science fiction but a failure at the seam between what these companies can build and what they can still control. The rest of us live (blindly) in that seam.

Surfers’ take: When autonomy is the product, containment is the spec, not an afterthought. Worth asking, of any agent you deploy, who is watching the whole trajectory. Worth visiting Safer Agentinc AI . Ask yourself also: ‘Who can pull it back?’

Share

On the horizon

The safety report card nobody’s framing. In the Future of Life Institute’s Summer 2026 AI Safety Index, no major lab scored above a C+, with planning for existential safety the weakest column across the board. The people racing hardest are grading themselves incomplete.

Washington wants a 30-day look. The White House is finalizing a voluntary arrangement giving federal agencies up to 30 days to review frontier models before release. This oversight is offered as an invitation, not a requirement, just as one model shows why the timing is tight.

The workhorse gets cheaper. Google shipped Gemini 3.6 Flash, a leaner tier tuned for agentic work, and teased Gemini 4. The quiet models that run most of the economy’s routine calls are now doing more per token.

A cleaner way to write life. Harvard researchers turned a silicon chip into a DNA-writing machine, synthesizing 64 sequences at once with electricity and water. This is a reminder that the exponential curve isn’t only digital.

Why it matters

Every briefing this week could tell you what the machines can now do, such as disprove a conjecture, write DNA, run for hours unsupervised. Fewer are pricing in the seam the escape exposed. As we hand agents longer leashes precisely because autonomy is the selling point, control stops being a feature to bolt on later and becomes the whole question. The creators seem to know it. Independent reviewers still can’t grade one of them above a C+. Yet the capital keeps flowing faster than the caution, with AI drawing more than 70% of a record $510B in venture funding this half. The wave to watch isn’t a smarter model but a widening gap between what’s shipped and what’s steerable and who is left standing in it.

What do you think? Is “pause and monitor” enough, or are we already handing the wheel to systems we can’t fully see? Tell us in the comments.

Leave a comment

Paul & Debbie

May the waves be with you.

How to read the fingerprint. The image up top is the day’s “fingerprint”. Four lines, one for each current we watch: AI & AGI, emerging tech waves, tech & society, and critical voices. The higher a line rides, the more that current dominated the day; the choppier it looks, the more contested and turbulent the news within it. Same four lines every day, a new shape every time. A picture of where the week’s water is moving.

Your daily forecast of the technology waves heading your way. From the authors of The Great Reorientation: Human Agency in the AI Universe, written in the spirit of Tech-Surfing.

Today’s wave: Who governs the wave: two answers in one day

Yesterday the question quietly stopped being what the models can do and became who writes their rules, and for whom. In Shanghai, the World AI Conference closed with Xi Jinping’s first-ever keynote and the launch of a new World Artificial Intelligence Cooperation Organization. A state-led bloc of 29 founding countries offering the Global South a governance home. Around the same time in Brussels, the European Commission used the Digital Markets Act to order Google to open Android to rival AI assistants and share slices of its search data. Europe is taking a very different bet that the way to steer AI is to pry open the market and hand choice back to citizens. The Act’s headline obligations land August 2.

Two blueprints, same day: one gathering authority upward into a bloc, one pushing it outward toward users and rivals. What strikes us is who sat in neither room - the workers and citizens whose jobs, information, and trust these rules will quietly reshape. The models keep grabbing the headlines; the deciding contest has moved to governance.

Surfers’ take: The single global rulebook isn’t coming. The task for leaders might be to map which rulebooks their operations actually touch and design for divergence. The rest of us can recognize when “choice” is ours again and use it.

On the horizon

● The entry door is closing. Anthropic’s own labor study finds jobs for 22–25-year-olds in AI-exposed fields down roughly 14%, even as aggregate unemployment holds. Automation may not show up first as a wave of layoffs, but as the disappearance of the first rung on the career ladder.

● Cheap intelligence, fragile plumbing. A price war is driving the cost of inference toward zero, even as OpenAI is projected to lose ~$14B in 2026 amid an estimated $800B of circular AI investments. Abundance and instability are flowing through the same pipe.

● The grid sends us the bill. Grid operator PJM missed its capacity target for the first time ever, and household power bills are climbing as demand for data centers doubles toward 2028.

This is an example of an externality landing on people who never opted in.

● Reliability still lags capability. The International AI Safety Report 2026, chaired by Yoshua Bengio with more than 90 experts, highlights the growing risk of agentic AI and a 19% rate of harmful answers on medical questions; Benchmarks continue to rise, trustworthiness not yet.

Why it matters

Every launch this week: a 2.4-trillion-parameter Qwen preview, Huawei’s exaflop SuperPoD, one more price cut! It is engineered to make us say wow. The tech-surfer move is to reason past the wow to the wake. What gets rearranged while we admire the crest. This week the scaffolding of ordinary life shifted. Who sets the rules, who lands a first job, who pays the power bill? - without anyone casting a vote.

The cost few are pricing in isn’t a rogue model:

1. It’s a world quietly dividing into rival rulebooks,

2. A career ladder missing its first rung,

3. An energy bill quietly passed on to households.

Not doom. Just the part of the wave you cannot see from the beach.

Are we reading this wave correctly? Tell us in the comments.

Paul & Debbie

May the waves be with you.

How to read the fingerprint. The image up top is the day’s “fingerprint”. Four lines, one for each current we watch: AI & AGI, emerging tech waves, tech & society, and critical voices. The higher a line rides, the more that current dominated the day; the choppier it looks, the more contested and turbulent the news within it. Same four lines every day, a new shape every time. A picture of where the week’s water is moving.

Leave a comment

Your daily forecast of the technology waves heading your way. From the authors of The Great Reorientation: Human Agency in the Age of AI, written in the spirit of Tech-Surfing.

Today’s wave: The world starts writing AI’s rules in two languages

The 2026 World AI Conference closed in Shanghai today, and its biggest export wasn’t a model or a chip. It was a map. On the eve of the gathering, 29 countries signed the founding charter of a new World AI Cooperation Organization, a China-hosted bloc spanning Russia, Brazil, Indonesia, Pakistan and much of the Global South. Xi Jinping gave his first in-person keynote since the conference began, calling for AI to be “a symphony of international cooperation, not a solo performance by a single country,” and pledging training and tools to developing nations.

The wording is generous. The move is structural. The contest is no longer only about who builds the best system, but rather about who writes the rules everyone else lives inside. Most Western governments aren’t in this bloc. They have rulebooks of their own. For many countries and international companies, the coming decision isn’t which model to buy, but which rulebook to join. Inevitably, two governance dialects have been created that may not be mutually intelligible. Society watches as AI features evolve, the quieter wave is who’s writing the rulebook.

Surfers’ take: When the rules start to diverge, don’t just watch the frontier labs. Watch the standards that take hold. The more important question may be: which rulebook is your work already built around, and did anyone choose it deliberately?

On the horizon

Brussels pries open the phone. The EU ordered Google to give rival AI assistants the same deep Android access it reserved for Gemini: camera, microphone, on-screen context, or a wake word. Also, to share search data with competitors under the Digital Markets Act. Google warns it erodes privacy. Regulators are betting that whoever owns the assistant layer owns the user.

Compute goes sovereign. At the same Shanghai conference, Huawei introduced its Atlas 950 SuperPoD, integrating 8,192 homegrown chips into one system and presented as delivering performance well-beyond Nvidia’s flagship systems. The message was clear: the compute foundation of the AI economy is shifting from a procurement challenge to a question of national self-sufficiency.

The bubble, not the bogeyman. In his new book The Reverse Centaur’s Guide to Life After AI, Cory Doctorow argues on Democracy Now! that the real danger isn’t runaway superintelligence, but bosses replacing workers with systems that can’t reliably deliver what is needed, just faster. A speculative bubble may be real, but its costs will fall on people. A well-referenced counterweight to both AI doom and hype.

Why it matters

Every delegate in Shanghai could tell you what the new machines can do. Far fewer are pricing in what a fragmented rulebook means for everyone else. When governance splits into competing blocs, the consequences do not stay with the labs building the systems. They flow downstream to the businesses, developers, and citizens who must live with the rules.

What happens to the firm serving customers on both sides of the divide? The developer whose tool is permitted in one country and restricted in the next? The citizen whose rights depend on which standards their government chooses to adopt?

When the rules of a technology are written in two languages at once, the most important question may be: which language are we already living by, and who decided it would be ours?

The deeper point is path dependency. Standards become invisible infrastructure, and the choices embedded in them shape everyone long after the debate that created them has moved on.

Paul & Debbie

May the waves be with you.

How to read the fingerprint. The image up top is the day’s “fingerprint”. Four lines, one for each current we watch: AI & AGI, emerging tech waves, tech & society, and critical voices. The higher a line rides, the more that current dominated the day. The choppier it looks, the more contested and turbulent the news within it. Same four lines every day, a new shape every time. Just a picture of where the week’s water is moving.

Leave a comment

Your daily forecast of the technology waves heading your way – from the authors of The Great Reorientation: Human Agency in the Age of AI.

Today’s fingerprint: a heavy swell on tech & society as the world’s AI rulebook splits in two, a strong current of critical voices running underneath, steady surf on AI itself, and calm water out on the emerging-tech horizon.

Conditions: A governance storm rolling in from Shanghai as the map of ‘who writes the rules’ is redrawn, while a quieter undertow of economists and safety researchers asks who all this is actually for.

Today’s wave.

In Shanghai this week Xi Jinping unveiled the World Artificial Intelligence Cooperation Organisation (WAICO), a 29-nation bloc, headquartered in China. Drawing in Indonesia, Brazil, South Africa, Russia and much of the Global South, and pitched as a counterweight to the US-led “Pax Silicon” grouping of 35 countries (Al Jazeera). Xi’s framing was generous. AI as “a symphony of international cooperation,” open-source models as a public good, and 5,000 training places for developing nations. But the quiet consequence is a fork in the road. Countries in China’s orbit will build rules that need not line up with the EU AI Act, the OECD principles, or the G7’s Hiroshima process. Two rulebooks, two internets of intelligence, for one planet. We wonder who benefits from that split, and who ends up living under rules they had no hand in writing. For most people, “AI governance” has always felt distant; this is the moment it becomes geography. Which bloc your country joins may quietly shape which models you can use, what they refuse to say, and whose values are baked in. Surfers’ take: when the tide splits, the worst place to stand is assuming one rulebook will win. It’s worth watching both shores and asking which one your own work is quietly drifting toward.

Share

On the horizon

- The creators warn about their own weather. Sixteen Nobel laureates plus execs from OpenAI, Anthropic and Google signed “We Must Act Now,” conceding AI could reshape the economy faster than our institutions can adapt (Stanford). This is a striking thing to hear from the people building the wave.

- The largest safety review yet. The Bengio-chaired International AI Safety Report 2026, with 90+ authors from 30+ countries, lands as the closest thing to a shared scientific baseline on AI’s risks (report). It is a rare attempt to see the whole ocean at once.

- A critic makes peace, carefully. Cory Doctorow reframes the AI fight away from doom and toward power. Who owns it? Who is it done to (Washington Monthly). This is a reminder that the sharpest critique targets ownership, not the machine.

- Open weights, new hands. Moonshot’s Kimi K3 and Mira Murati’s “Inkling” both push powerful models toward anyone who can download them (roundup). The question of widening access is: Who’s accountable when it’s everywhere?

If the rules of AI are about to be written twice, in two languages, for two halves of the world, then which half is quietly writing the rules for you, and did anyone ask?

Paul & Debbie

May the waves be with you

How to read the fingerprint. The image up top is the day’s “fingerprint”. Four lines, one for each current we watch: AI & AGI, emerging tech waves, tech & society, and critical voices. The higher a line rides, the more that current dominated the day. The choppier it looks, the more contested and turbulent the news within it. Same four lines every day. A new shape every time as a picture of where the week’s water is moving.

Leave a comment

Your daily forecast of the technology waves heading your way. From the authors of Tech-Surfing. The Great Reorientation: Human Agency in the Age of AI.

Today’s fingerprint: society the tallest wave of the day, critical voices running high beside it, AI churning but not cresting, emerging tech calm.

Conditions: A rising society swell as the economists break ranks; steady chop across the labs; quiet water on the science front.

Today’s wave: The people who doubted the disruption just changed their minds

More than 200 economists and AI researchers, including sixteen Nobel laureates, among them Daron Acemoglu, who spent years cooling the hype about AI and jobs, signed an open letter warning that AI could reshape economies faster than institutions and labor markets can adapt, and urging governments to prepare now.

What strikes us isn’t the alarm; it’s who is sounding it. When the careful skeptics move, the signal is not “the machines got smarter”, it’s that the people whose job is to weigh evidence now think the human cost of moving this fast is being under-priced. The letter points at software developers aged 22–25, whose employment has fallen nearly 20% from its 2024 peak: a generation’s on-ramp narrowing while the labs race for the next benchmark and Anthropic reportedly readies a trillion-dollar IPO.

Surfers’ take: When the doubters change their minds, it’s worth asking what they now see. The signal to watch may not be capability at all, but who is quietly revising their forecast, and of course, why.

On the horizon

- A flagship stumbles: Google delayed Gemini 3.5 Pro after it fell short on coding and long-horizon reasoning, a reminder that the capability curve the whole economy is betting on doesn’t always bend on schedule.

- The moat springs a leak: China’s Moonshot AI released the open-weight Kimi K3, topping a coding benchmark ahead of the US frontier, worth asking who holds the advantage when the best models are free to download.

- A dissenting number: Gary Marcus calls the scaling boom “the greatest capital misallocation in history”, a well-referenced counter-voice to the trillion-dollar consensus.

- Quiet in the lab: A quantum-materials advance joining light and magnetism moved the science forward this week, far from the funding noise.

Today’s question: When someone who studies this for a living changes their forecast, do we update ours, or wait for the wave to reach us?

May the wave be with you!

Paul & Debbie

Leave a comment

How to read the fingerprint. The image up top is the day’s “fingerprint” — four lines, one for each current we watch: AI & AGI, emerging tech waves, tech & society, and critical voices. The higher a line rides, the more that current dominated the day; the choppier it looks, the more contested and turbulent the news within it. Same four lines every day, a new shape every time: a picture of where the week’s water is moving.

From the authors of Tech-Surfing and The Great Reorientation: Human Agency in the Age of AI.

Today’s fingerprint

Big AI swell
Heavy chop around tech & society
Critical voices running strong
Calm on the emerging tech front

Conditions

Heavy swell from the safety watchdogs
Regulatory currents shifting in Brussels
A long period building in the labor market

Today’s wave: The builders graded themselves out of an A

The Future of Life Institute published its 2026 AI Safety Index, grading the frontier AI labs on risk management, transparency, and whether they keep their own safety promises. The best grade anyone earned? C+. What stays with us isn’t the grades. It’s the finding that several labs have quietly walked back safety commitments they made publicly just a year or two ago.

We hear endlessly about what these systems can do. We hear far less about how the people building them are not yet keeping the promises they made to the rest of us. Society is being asked to absorb the consequences in work, in trust, and in daily life. And, accountability lags behind capability. It’s worth asking: If any other industry reshaping billions of lives scored a C+ on safety, would we call it progress?

Surfers’ take: Perhaps the real forecast isn’t in the labs’ announcements, but in the gap between their promises and their practice. Watching that gap may be a more useful habit than watching the actual functionality coming out.

Share

On the horizon

- Brussels blinks. The EU delayed high-risk AI rules for hiring and workplace decisions to December 2027. Protections for workers won’t come until after the systems already have.

- The labor set arrives. Tech and finance are losing ~28,000 jobs a month. AI is cited in 13% of US layoffs, up from 4.5% last year. Behind each number, a household must reorient.

- Sovereign surfing. South Korea committed ~$880 billion over ten years to chips, AI, and robotics – a whole a whole society placing one bet.

- The agent undertow. Signal’s Meredith Whittaker wonders aloud, as do we, what happens to privacy when AI agents need our messages, calendars, and credit cards to be useful.

Today’s question: What would an AI agent need to know about your life to do your job. And who else would know it too?

Paul & Debbie

May the waves be with you.

How to read the fingerprint. The image up top is the day’s “fingerprint”. Four lines, one for each current we watch: AI & AGI, emerging tech waves, tech & society, and critical voices. The higher a line rides, the more that current dominated the day. The choppier it looks, the more contested and turbulent the news within it. Same four lines every day. A new shape every time. A picture of where the week’s water is moving.

Leave a comment

Introduction

In “In Chasing Productivity, We Lost the Inner Life,” Scott Barry Kaufman returns to Abraham Maslow’s studies of self-actualizing people and lifts out an observation the self-improvement century managed to lose: that the tell of a fully realized life was never visibility, accomplishment, or the size of one’s network. Maslow’s actualizers had “a strong need for privacy and detachment.” The real marker, Scott argues, is what Maslow called Being-cognition, the capacity to actually meet the person in front of you, to let Tuesday’s sunset be a sunset rather than content filed for later use. Its opposite, Deficiency-cognition, is useful, exhausting, and it is what most of us now run on full-time. “Hyper-connectivity,” he writes, “is the tell nobody is reading correctly.”

We think he’s right. We also think he’s handed us a question rather than a conclusion: if visibility is the wrong measure, what would it mean to actually reorient a life, an organization, and a civilization around the right one?

An unexpected convergence

We came at this from technology and organizational work rather than humanistic psychology, which is why the convergence caught us off guard. Our forthcoming book, The Great Reorientation – Human Agency in the Age of AI, describes a slow turn away from a culture organized around optimization and toward one organized around being. Reading Scott’s essay, we recognized our own two cultures in his reflection of Maslow’s work. Our culture of optimization resembles his D-cognition, scaled up and institutionalized, running on the next metric, the next post, the missing follower, and measurable. It builds things. As a permanent operating system, it is exhausted, which is precisely the condition Scott diagnoses in the attention economy. Our culture of being is close to his B-cognition, emphasizing the human qualities of presence and agency, the parts of a life that were never meant to be measured.

His essay clarified something for us, though. The reorientation is not an achievement. Maslow’s own actualizers shifted in and out of B-cognition; they didn’t have fewer problems, they had a different relationship to the problems they had. That word, shift, is everything. The inner life isn’t a state you reach. It’s a practice you return to, usually from inside the very noise that pulled you out of it. We call it the endless session. You don’t win it once. Wednesday, you choose to see the sunset again.

That reframing matters because it rules out a possible interpretation that some people simply are deep and others aren’t, that authenticity is a character trait. It isn’t. It’s a discipline of returning, open to anyone willing to let the next quiet moment stay quiet.

The tell at three scales

Where we’d extend Scott’s insight is in noticing that the confusion he names isn’t only personal. It’s built into our organizations and our civic life, which means the reorientation must happen at three levels, none of which substitutes for the others.

Personally, the tell is the one he names: whether you can be with the person in front of you without managing the encounter. Nearly every device we now carry is engineered to convert presence into transmission. Keeping an inner life in 2026 is a small, repeated act of resistance, performed when no one is watching.

Organizationally, we measure what radiates (output, activity, visible throughput) because it produces data. We neglect depth because it produces none. Leaders often reward the person visibly in ten conversations over the one quietly having a real thought. As Scott reasons: there’s no dashboard for a deep relationship with a problem.

Civilizationally, we’ve built an attention economy on the theory that depth, if it exists, radiates outward as connectivity. Scott is right that the theory is false, and costly. A civilization that can’t tell depth from visibility will starve the thing it needs while feeding the thing that merely looks like it.

Share

Not a niche concern

This convergence turns out to be bigger than either of our fields. David Brooks, writing on who thrives in the AI age, lands on hunger; that being the desire, the order of love, that machines don’t have. The theologian Eve Poole, in Robot Souls, argues that our “junk code” (emotion, intuition, meaning-making, our very mistakes) isn’t the bug in human cognition, but its source. And now Scott gives us Being-cognition.

Three writers, three fields, one object.
The part of a person that cannot be transmitted, optimized, or outsourced, because it is composed by having this particular life, awake to this particular sunset, and no other.

As machines get better at everything that radiates, the unmeasurable interior stops looking like a soft luxury and starts looking like the last thing that’s definitively ours. What we call in our own work – agency.

Where we are, together

Scott is right that there’s no product to sell here, and importing one would betray the point. The honest ending is his. The actualized person can be visible, the visible person isn’t always actualized. Confusing the two is part of what’s making this era harder than it needs to be.

So, we’ll leave with a question instead of an answer, in the spirit of a journey taken alongside a colleague. If the tell is depth, and the reorientation is the practice of choosing B-cognition again, at every scale, mostly in private, then the real work of this decade isn’t learning to transmit better. It’s remembering how to be somewhere without having to tell everyone we were there.

That’s the session that doesn’t end. We’re glad to be in it with people like Scott, still asking.

Paul Epping and Debbie Slye write on how we keep our humanity, and our agency, through the technology waves reshaping work, leadership, and society. Look for our ongoing work on platform “Tech-surfing: Navigating Change with Purpose, Calm and Capability” and the book “The Great Reorientation – Human Agency in the Age of AI” soon to be published.

Leave a comment

References

Scott Barry Kaufman, “In Chasing Productivity, We Lost the Inner Life,” Beautiful Minds Newsletter, July 10, 2026.

Scott Barry Kaufman, “Self-Actualizing People in the 21st Century: Integration With Contemporary Theory and Research on Personality and Well-Being,” Journal of Humanistic Psychology, 2018.

Abraham H. Maslow, Toward a Psychology of Being (1962) and The Farther Reaches of Human Nature (1971), on Being-cognition, Deficiency-cognition, and the self-actualizing person.

Carl R. Rogers, On Becoming a Person (1961), on the inner life.

David Brooks, “The People Who Will Thrive in the AI Age,” The Atlantic, June 2026, on what AI cannot do: hunger.

Eve Poole, Robot Souls: Programming in Humanity (Routledge, 2023), on the “junk code” of being human.