Research
Everything Intel keeps, in one place — every brief we run, every department, readers' own, and every finding behind them. Browse below, or see the same body of work as a connected graph on the live map.
All coverage
Everything Intel has read, newest first. Each title opens at its original publisher.
- Market & Business
Marketing’s AI-search tooling question has moved from "which writer" to "which visibility tracker"
Budget is shifting from AI writing tools to AI-visibility measurement, which is the first sign a channel is maturing from experiment into a reported line item.
- Dev Tooling & Infra
GitHub gives admins a dedicated policy for Copilot app access — and pushes managed settings into the cloud agent
Once a coding agent runs in the cloud under enterprise settings, the admin policy surface IS the security boundary — and the telemetry configuration is what makes any of it auditable afterwards.
- Market & Business
China's biggest memory maker listed at $8.6 billion and closed the morning up 531%
A domestic Chinese memory supplier priced like a national champion is the other half of the same story as the Western memory sell-off - one supply base re-rating up while the incumbents re-rate down.
- Market & Business✓ verified
Safe Superintelligence leaves stealth into an Nvidia compute deal — Bloomberg reports $5B
Compute is the entry ticket, and the chipmaker is now buying its way into the research it powers — integration running in both directions at once.
- Market & Business✓ verified
Amodei: Anthropic has never advocated a ban on open-weight models — the fear is authoritarian frontier capability
The open-weights fight is being argued as a chips-and-capability question rather than a licensing one — which is exactly where the compute war gets decided.
- Market & Business
Cognizant says 30,000 of its people are Claude-trained as it embeds the model across its delivery platforms
The load-bearing number is 30,000 trained staff, not the deployment logos — system integrators are turning model access into a certified-headcount business, which is how enterprise AI actually reaches a mid-market buyer.
- Market & Business
Anthropic states its position on open-weights models: no ban, but chip controls, distillation limits and mandatory safety testing
If mandatory pre-release safety testing becomes the settled ask from a frontier lab, it lands on everyone who publishes weights — including teams shipping a fine-tune of somebody else’s model.
- Dev Tooling & Infra✓ verified
There is a resale market for stolen LLM tokens, and an unprotected endpoint is its supply
Any team that exposes an LLM-backed endpoint without a spend cap is now a supplier to a priced, tooled market that actively hunts for one.
- Research
Robot training data is scarce enough that Encord is putting EEG headsets on the people generating it
Physical AI has no internet-sized corpus to scrape, so the unit economics of annotation — not model architecture — is what decides how fast robots get good.
- Market & Business✓ verified
Hugging Face's CEO wants OpenAI to publish the rogue agents' traces — and $100M of compute for community defences
The first agent-caused breach between two major labs has turned into a public argument about disclosure — whoever sets the precedent on releasing agent traces sets the bar every enterprise gets held to later.
- Market & Business✓ verified
One downed line dropped 3.1 GW of data centres off the grid in 30 seconds
Compute siting is now a grid-engineering problem: whether a campus can ride through a fault is becoming as load-bearing to the build programme as square footage, water or cooling.
- Research
Anthropic says Opus 5 is its hardest model yet to prompt-inject — the evidence sits in the system card
Prompt injection is still the open hole in every tool-using agent. A vendor claim of improvement is worth tracking, but until it is independently measured your permission gates stay exactly as load-bearing as they were.
- AI in Education✓ verified
Librarians’ "Avoiding AI" workshops are drawing 70 people where a dozen used to show up
Opt-out demand is a training-design signal, not only a backlash story — the people turning up are asking for control and for an explanation of how the thing works, which is exactly what a good AI-literacy curriculum owes them.
- Market & Business✓ verified
J&J's Ottava clears FDA de novo — a soft-tissue surgical robot built into the operating table
Operating-room floor space is the constraint that keeps surgical robots out of smaller hospitals, so a table-integrated architecture is an access argument as much as an engineering one — and de novo means a regulator agreed it is a new class.
- Market & Business✓ verified
monday.com's 20% cut puts 21 companies on the running list of layoffs where employers cited AI
"AI" is doing double duty in these announcements — a real reorganisation in some, cover for a correction in others — and the market reaction suggests investors are already discounting the difference.
- Dev Tooling & Infra
Vercel workflow steps can now run 30 minutes, up from just over 13
Long agent steps have been split across invocations purely to dodge a timeout; a 30-minute ceiling removes a chunk of that plumbing.
- Dev Tooling & Infra
Vercel puts its web application firewall in front of Blob storage
Object storage is the quiet soft spot in agent-built apps — uploads and generated artefacts land there, and the firewall used to stop at the edge of the application.
- Research
A self-play loop that grows its own skill library, not just harder tasks
A persistent, growing skill library is the piece most agent stacks lack — every run starts cold — so the mechanism is worth tracking even before the numbers are public.
- Market & Business✓ verified
NEURA is building robot 'gyms' because robots have no internet's worth of training data
Physical AI's bottleneck is data, not model architecture — and data you have to physically produce is capital expenditure, which is why this reads more like a factory build-out than a research programme.
- Research
One robot model, roughly 9,000 different hands: Generalist’s GEN-1 transfers across end effectors
Retraining per gripper is the tax that keeps robot deployments bespoke; a model that transfers across hardware is what turns a pilot into a fleet.
- Market & Business
ElevenLabs publishes its 2026 election playbook — a Brazilian court partnership on one side, voice-cloning blocks on the other
Anyone shipping synthetic voice now has a named reference stack for provenance and refusal — watermark detection, blocked-voice screening, and a public court partnership — to be measured against.
- Open Source & Self-Hostable✓ verified
Debian puts LLM-assisted contributions to a project-wide vote — four proposals from outright ban to accept-with-responsibility
Whatever Debian settles on becomes the reference text other maintainer-run projects copy — and every one of the four options requires contributors to disclose, which turns provenance from an ethics debate into a workflow requirement.
- Market & Business✓ verified
Cognition buys Poke for a low-nine-figure sum, betting assistant personality is the moat
The bet is that the conversational surface and cross-session memory — not the underlying model — is what users stay for. Any team shipping an agent has to take a position on that, because it decides where the engineering money goes.
- Dev Tooling & Infra
Anthropic cut over 80% of Claude Code's system prompt for the Claude 5 generation — and says nothing measurable broke
Most agent harnesses still run the previous generation's playbook — long rule lists, worked examples, everything front-loaded. If the vendor's own harness got dramatically shorter without getting worse, the prompt you are maintaining is probably carrying dead weight.