A Microsoft pilot shipped verified fixes, a migration test needed an independent verifier, and enterprise buyers want one testing platform. Verification is where agent coding now stalls.
GitHub's session protocol, Google AI Studio's instant Android builder and a filesystem that cuts checkouts from 36 minutes to seconds. The dev loop is being rebuilt for agents.
Gemini's Task mode, Muse in Australia and Gemini 4 Hydrogen for robotics all ship the same idea: the agent acts, with your approval, and tells the business it is an AI.
Adobe promises choice over when to use AI, Google gives away avatars restricted to your own likeness, and Muse asks before it acts. Control is the pitch now.
Cerebras splits prefill from decode, AMD caches the history agents keep re-sending, and attention costs dominate long context. The inference stack is being rebuilt around agent traffic.
Three new studies point the same way: how an agent calls tools matters more than which model does the calling. The cheapest wins come from the interface, not the model.
Agent pipelines lose to a plain baseline on claim support, ignore primary evidence when a false claim repeats, and refuse badly when documents conflict. The same batch of papers shows what repairs each failure.
Four new evaluations find the same gap from different directions: models stumble on culture and language outside the English-language mainstream, and the tests meant to catch it are not catching it.
Isomorphic Labs raised $5 billion to design medicines, a Microsoft diagnosis agent beat human generalists, and one researcher runs the whole pipeline on Claude Code.
A Databricks revenue update, a SemiAnalysis spending breakdown, a survey of 150 IT leaders, and a venture firm's infrastructure thesis all point the same way: the buying has started, and the blockers are access, identity, and proof.
Four new systems attack the same problem from opposite ends: mark model output so it can be checked later, and detect synthetic media fast enough to run everywhere.
New work on debate-based rewards, self-training with value-guided search, and ensemble reinforcement learning points at the same shift: capability now comes from how models are trained after pretraining, and the critic in the loop needs its own checks.
Quantization, codebook sharing and smarter routing each cut memory or latency by a measured amount. The numbers are the authors' own, and none of it is peer reviewed yet.
Five new benchmarks measure long-horizon work, memory governance, visual evidence and exploration. The best models lead by wide margins — and the tests say short evaluations mislead.
Cambridge Associates says the case for equity long/short funds survives only where managers show real stock-picking skill. RBC BlueBay argues today's securitized credit is not the pre-crisis version. Meketa says the risks no longer sit in separate boxes.
The bank's 2027 outlooks put US growth above the 2.2% consensus and UK growth above 1.2%, driven by AI investment. The same outlooks expect rate hikes before any cuts.
Hyperscaler spending is projected at about $3.8 trillion from 2026 through 2028, with credit funding more than half. UBS puts AI capital spending at $1 trillion this year alone. The binding limits are physical and political.
The rate cycle is fragmenting. Hungary held, Poland is disinflating, New Zealand's central bank is still tightening, and the ECB is refining how it talks to markets.
A Cleveland Fed president argues the central bank should decide in advance when it would step into the $32 trillion Treasury market. New Fed research shows why the plan could fail: banks fear the market reaction to borrowing.
BlackRock's fixed-income chief argues income now does the heavy lifting in a portfolio, and Vanguard is cutting credit risk while expecting two more rate hikes. The catch: households have barely moved.
Morgan Stanley sees Korea's market passing 10,000 and China's frontier industries adding $10 trillion in market value by 2030. The latest data from China tells a slower story.
A phage cocktail that covers most drug-resistant Klebsiella, a new cholera phage, surging drug-resistant TB in Brazil and a returning pneumococcal strain in Catalonia.
Ryan Jeffers stayed with the Twins on a $23.15 million qualifying offer while Morejon, Rogers, Suzuki and Lowe turned theirs down. The White Sox, meanwhile, paid $52.5 million for certainty.
Effective November 17, the National Institute of Mental Health rescinded its autism and clinical-research data-sharing notices. Applicants now follow the NIH-wide data management policy.
Fibre-optic sensors tracked an Icelandic eruption, a new satellite method sharpens heat maps — and open geodetic data cannot detect magma building under 64% of the most dangerous populated volcanoes.