The agents wrote 200 services. Nobody reviewed a line.
AWS's emulator of 200-plus services was written entirely by agents with no human code review. The companies selling the verification layer are growing tenfold in a year.
Updated 2026-11-13 12:00 UTC
28 stories, 139 Manifests cited.
Artificial Intelligence
Agents don't need bigger models. They need somewhere to put the conversation.
Multi-turn agent workflows thrash the KV cache, and this week's papers attack it from every side: scheduling, compression, fleet-wide coordination and even removing the memory from the GPU.
The data platform is now the agent's office
Block built agents a chat room. Databricks put a governed agent in Slack. Constellation is building a nuclear engineering assistant on the same stack. The agent moved into the data platform.
The score says the model works. The work says otherwise.
Models keep most of their accuracy when the nouns are deleted. Judges miss forbidden writes. A week of evaluation research says the score is the least reliable part of the pipeline.
The attacker stopped hacking the app. Now they hack the agent.
This week's security research reads like a catalogue of agent-specific attacks — a 94x API amplification, a $20 jailbreak, a backdoor in ten neurons — and the defences are only just catching up.
Agents are doing the maths now. The proof is the product.
A general-purpose model solved all 12 Putnam problems in Lean. A cancer-sequencing agent beats fixed thresholds with an audit trail. Science is where agents prove themselves, literally.
When agents fail, the best auditor is the agent itself
POIROT beats centralised LLM evaluators by up to 50 points at blaming the right agent. The catch: when agents talk to each other about the failure, they converge on the wrong culprit.
Agent memory is now a product. It is also a hole.
GitHub's Copilot now remembers your repo's conventions, with a 7% merge-rate lift. The same week, researchers showed memory systems scoring 1-3% on dependency reasoning — and a live command channel hidden in an agent's memory.
Your agent's answers are only as honest as its context
The week's RAG papers and data-platform posts agree on one thing: the model is not the risk anymore. The context it acts on is.
The forecast is better than ever. The physics is not.
Google DeepMind outperformed the National Hurricane Center on storm track this season. A storm-resolving AI runs 50 times more efficiently than the physics model. And a new framework exists because the models hallucinate physics.
The voice agent that answers the phone, and the voice that lies on it
One credit union's agent handled 598,000 calls in 11 months. The same week, prosecutors indicted a $129M scheme built on cloned CEO voices. The phone is now an AI battleground.
The small model that can't do the job can still grade it
A System-1 judge costs a sixteenth of a frontier reasoning model and grades about as well. But the best model on accuracy ranked seventh on reliability, and some judges flip half their verdicts under pressure.
Agent-written code survives review as well as yours does.
Across 2,794 merged pull requests, AI-written code was not modified sooner than human code. The work that remains is review, ownership and detection — not the writing.
Research & Science
The chemistry bench is learning to run itself
A self-driving electrochemistry platform, software that beats experts at tuning spectral models, and 3D-printed reactors that hold up at 160 °C — plus catalysts that switch themselves off before they overheat.
Finance & Investing
Three risk monitors, one message: the system is bending
Bank vulnerability gauges are rising, insolvencies are expected to climb, and Germany's financial watchdog counts private credit and cyber threats among the risks. Resilience is intact. The direction is not reassuring.
Six companies revised guidance in a day. Four paired it with a promise.
The reasons differ — household budgets, a weak Brazil, old assortment mistakes — but the shape repeats: a lower bar for this year, and something to look at instead.
ING's 2027 FX outlook: a quiet year with fat tails, and a dollar that falls late
Rate gaps, not geopolitics, drive currencies next year, ING argues. The yen looks best-placed, the peso most exposed, and a tech selloff is the risk that matters most.
Crypto's split quarter: the exchange shrinks, the treasuries profit
Gemini lost $58.6 million as trading volumes fell by nearly three-quarters. Upexi made $89.1 million, almost entirely from unrealised gains on its Solana holdings. Same quarter, different business.
Central banks have stopped watching AI from the sidelines
The Czech National Bank says its AI model forecasts GDP almost as well as its economists. The Fed's own research says AI's effect on jobs and productivity is still small. The gap between those two facts is the story.
The AI buildout's bill is arriving in three places at once
Apollo says software's 2028 maturity wall is barely shrinking. Morgan Stanley puts Asia's grid bill at $1 trillion by 2030. The buildout is not slowing down; it is getting more expensive to fund.
Health & Biotech
The drug evidence disagrees with itself
Drug effects don't reproduce across datasets, trials contradict each other, and genetics predicts only a little of antidepressant response — but automated auditing is catching up.
Cancer's immune guards are the new drug target
In glioblastoma, prostate cancer, neuroblastoma and breast cancer, the cells that shield tumours from immunotherapy are coming into focus — and so are ways to remove them.
Sports
Baseball's $100,000 deadline: protect your prospects by Tuesday or lose them
All 30 teams must set their 40-man rosters by Tuesday at 6 p.m. ET. Eight Top 100 prospects are exposed, the Rule 5 draft follows on Dec. 9, and the general managers are already moving.
One month in, three NBA teams are already rethinking their plans
Indiana is shopping for mid-tier help under a tax crunch, New York's starting five is not working, and New Orleans just had the worst three-game stretch in franchise history. None of this was the plan.
Federal Government
ICE says 2026 was its biggest enforcement year ever
The agency reports 483,200 deportations and more than 491,000 arrests in fiscal 2026, both above the combined totals of the three prior full fiscal years. Two arrests this week show the pace.
Federal enforcers spent the week taking down the access brokers
A botnet that sold ransomware gangs their way in, dark-web drug markets, and two insider-trading cases built on stolen filing access. The week's cases all hit the middlemen.
State Legislatures & Executives
States took their own climate pitch to COP31
Minnesota brought 40 laws and $1 billion, California brought wildfire expertise and a Portugal deal, and Massachusetts brought a cross-border wind plan. The states are doing climate diplomacy themselves.
Energy & Climate
Every climate fix trades one outcome for another
Removing carbon earlier or later, spraying aerosols at the equator or the poles, counting dust or ignoring it — the week's modelling shows each choice buys one climate outcome at the price of another.