Milestone: Weāve now crossed 1,000 subscribers in four weeks. Thanks for reading, sharing, and thinking with me. Letās keep the AI signal sharp and real.
This week: agents, geopolitics, AI loops, and why writing still matters.
š§ AI Culture, Hype, and the Human Operating System
š§¾ McKinsey wonāt die. Because humans donāt change that fast.
Despite the cries of AI replacing consultants, the role of firms like McKinsey isnāt about knowledge. Itās about narrative cover. In high-stakes environments, strategy is theatre. Itās about defensibility, status, and shared risk, not optimal answers. Until AI can absorb blame and signal credibility, execs wonāt replace the humans they can lean on.
Read post ā
š Reddit Gets It First.
The Reddit crowd gets real about the AI market faster than corporate channels do. This post reflects on where real insight shows up, why nuance can look like negativity, and what early signals of actual value often feel like in the noise.
Read post ā
š©ŗ GPT-4 judges itself. No surprise, it passes.
A new paper claims GPT-4 can outperform doctors in diagnosing and treating patients - but it also lets GPT-4 assess its own performance. I pair this with a second paper exposing the flaws of evaluating AI with human tests. The result? A deeper dive into āevaluation theatreā and why fluency shouldnāt be mistaken for reasoning.
My first Substack deeper dive:
Read full essay ā
šŖ The AI-police are barking up the wrong tree.
When someone said my writing āsounds like it was written by an advanced AI,ā they meant it as a compliment. This reflection explores how I use LLMs not to write but to sharpen - clarity, rhythm, transitions - and why tooling doesnāt erase voice. It can amplify it.
Read post ā
āļø Agentic Systems, Infrastructure Shifts & Real Use
š Agentic Web: the power shift nobody sees.
As agents take the reins, the web becomes less about UX and more about structured invocation. This post explores a conceptual paper about the āmid usersā of the future - autonomous agents - and how trust, discoverability, and selection are becoming the new SEO.
Read post ā
š¤ Magentic-UI: autonomy reimagined with humans at the centre.
Microsoft open-sourced a human-in-the-loop agent system. But what makes it radical isnāt the code - itās the philosophy. This system builds around human agency, supporting interaction modes like co-planning and memory reuse. A welcome antidote to full-autonomy fantasies.
Read post ā
šļø A blueprint for process-aware agents.
Instead of throwing agents at broken workflows, this paper starts with business process design. It reframes workflows around outcomes, objects, and declarative logic - before layering in autonomy. Itās not just smarter automation, itās a provocation for how we think about structure.
Read post ā
š§ Metaās Agent may be your next context trap.
Metaās announcement of āpersonal superintelligenceā converges with a survey on self-evolving agents. What emerges is the notion of contextual lock-in: the tighter your agent aligns to you, the harder it is to leave. Itās not coercive. Itās sticky. And it reshapes what AI loyalty even means.
Read post ā
š Geopolitics, Power & AI Governance
šØš³ Chinaās AI plan isnāt soft - itās strategic.
While the US leads with stack control, Chinaās AI governance pitch uses multilateralism and infrastructure diplomacy. This post dissects how sovereignty, influence, and technical framing show up behind velvet-glove language. The world must choose not just stacks - but values.
Read post ā
š The model moat is shrinking.
Menlo Venturesā report: capex is rising, model advantage is falling, and orchestration layers are where the real leverage now lies. This post synthesises those trends - shifting from magical thinking to pragmatic infrastructure.
Read post ā
āļø Frontier drama: Claude vs GPT.
OpenAI was caught ābenchmarkingā Claude. Anthropic cut them off. Not much to add. Just a reminder that the frontier isnāt just technical - itās political.
Read post ā
š Loops, Woozles, and Evaluation Theatre
š§© TreeReader and the risk of synthetic smoothness.
A tool that summarises academic papers sounds useful - until you realise many of those papers were written by LLMs to begin with. This post explores the feedback loop of simplification, the erosion of friction, and the importance of reading things that still resist understanding.
Read post ā
š¾ The Woozle Effect, AI edition.
LLMs donāt always repeat truth. They repeat whatās repeated. Over time, that becomes its own kind of truth. Footprints in the snow become āevidence.ā A whimsical reflection with serious implications for epistemology.
Read post ā
š Agents fail when red-teamed. Badly.
In a massive red-teaming study, nearly every agent broke within 10ā100 queries. The results are concerning: leaked data, criminal suggestions, policy breaches. Weāre building autonomy before weāve built oversight.
Read post ā
āļø Writing, Prototypes & Product Judgment
š ļø Vibe-coded prototypes ā clarity.
Googleās shift away from āwriting cultureā prompts this reflection on what writing actually does. It isnāt theatre - itās where contradictions get surfaced. Writing is judgment. Itās where bad ideas die before they become bad software.
Read post ā
š Lighter Bits
š¤ Perplexity recommends your agent log into your bank.
AI as personal finance assistant - or fraud risk? Sometimes the absurdity just writes itself.
Read post ā
š¦øāāļø One feature. One fix. One hero.
Some days, the fantasy of product workā¦
Read post ā
ā
CAPTCHA overthinkers, unite.
A checkbox becomes a mirror.
Read post ā
š SEO hack or ouroboros?
ChatGPT conversations indexed by Google. Google indexes ChatGPT. AI eats its own tail.
Read post ā
šļø New Episode: The Unhype AI Podcast
Episode 5: AI, Power, Hype, and the Illusion of Progress
and I dig into whoās actually winning with AI and at what cost.š§ US strategy as doctrine, not direction
š§± China pushing back on hype
šŖšŗ EU stuck in regulatory limbo
š Big Techās real power map
š§ Agents, structure, and what augmented orgs could look like
š§ Plus: child-safe AI, AI companions, promptfluencers, and the myth of trillion-agent futures
Satirical, grounded, and maybe our most urgent yet.
š§ Listen on Spotify
šŗ Watch on YouTube
Thanks for reading.
If something here made you pause, thatās the point.
And if it made you laugh, even better - because weāre going to need humour if weāre going to make it through the hype.
Until next week,
Stuart x
Seen something here you want to unpack with your leadership team?
I run briefings and workshops tailored precisely for that.
You can also work with me in a fractional advisory, product, or AI strategy capacity - bringing ongoing insight, strategic framing, and challenge at the leadership level.
If your leadership team needs clarity - not theatre - letās talk
Congrats on the milestone, Stuart!