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.
- Agent Frameworks & Orchestration
Nvidia research puts the agent harness, not the model, at the centre of the result
If the harness carries the outcome, agent quality is engineering you own rather than a model you rent, and it is the half you can actually iterate on.
- Agent Frameworks & Orchestration✓ verified
Nvidia's agent wrapper took Claude Opus 5 from 30% to 100% on a reasoning benchmark
If the software around the model — memory, supervision, tools — is worth seventy points on a long task, then agent quality is mostly engineering you control rather than a model you wait for.
- Agent Frameworks & Orchestration
Microsoft's Agent Framework adds agent-generated interfaces and a fatal-middleware signal
An agent that renders its own interface, and a middleware failure that actually stops the run, are both about giving the operator somewhere to stand.
- Open Source & Self-Hostable
llama.cpp starts a stable release channel with v0.2.0
Packagers and anyone shipping llama.cpp inside a product finally have a version to pin that does not move with every commit.
- Frontier Models
GPT-5.6 Sol list pricing drops to $4 and $20 per million tokens
Frontier-tier pricing is now moving down between releases rather than only at them, which is worth re-running the unit-cost maths for on anything built against the previous number.
- Research
Google builds a tool to pick candidate biomarkers out of wearable sensor data
Wearables produce far more signal than anyone can chase, so the useful product is the shortlist rather than the stream.
- Dev Tooling & Infra
Claude Security's vulnerability scanning moves onto Anthropic's most cyber-capable model
A frontier model reaching security teams as a product switch, rather than as an API they have to wire up themselves, is how capability actually lands in a security org.
- Market & Business✓ verified
US futures regulator opens public comment on trading compute as a derivative
Once compute is a listed, hedgeable contract, a large training or inference bill becomes something you can price and forward-buy instead of just absorb.
- Research
An MIT excavator interface lets a first-day operator work like a veteran
Where skilled-operator shortages bite hardest, closing the training gap at the interface is a faster fix than replacing the operator.
- Market & Business
Alibaba's AI build-out lifts capital spending 75% as quarterly profit falls 76%
One of the clearest public readings yet of what the AI build-out costs a profitable incumbent in the quarter it is spending.
- Research
A measurement of how much speech-recognition progress is benchmark optimisation
Anyone choosing a speech model off a leaderboard is choosing off a number that may have been optimised for, and this is the correction factor.
- Research
Two scientists analysed 112 released Pentagon UAP videos and found the footage cannot establish speed
A worked example of a release that looks like disclosure and is not usable as evidence: the metadata that makes the measurement possible is exactly what was withheld.
- Market & Business
PlanRadar puts agents into construction admin — and they act without an approval step
Approval-free automation arriving in a liability-heavy trade is exactly where the question of who signs off gets decided in practice.
- Agent Frameworks & Orchestration✓ verified
n8n adds an Amazon Bedrock AgentCore node for multi-agent teams with persistent customer memory
Persistent per-customer memory is the feature that turns a workflow automation into something that behaves like an account team — and the fact that it now arrives as a node rather than a build is what puts it in reach of a non-engineer operator.
- Dev Tooling & Infra
VS Code 1.134 lets one window drive agent sessions running in another
The prompt timeline and cross-window session hosting are the first-class answer to the thing that actually breaks a long agent run — nobody can tell which prompt made which edit. Reviewing an agent session is becoming a supported workflow rather than a scroll.
- Market & Business✓ verified
Unitree’s IPO gives the humanoid sector its first public comparable
Public filings mean the humanoid sector finally has audited unit economics somebody outside the cap table can read — the first real check on the revenue claims the category has been valued on.
- Market & Business✓ verified
Stripe acquires OpenRouter, reportedly for $7.5B
If you route production traffic through OpenRouter, your model gateway is about to be owned by your payments provider — a concentration worth naming in a risk register now, while the deal is still closing.
- Market & Business✓ verified
Serve Robotics puts its sidewalk delivery robots on Grubhub
Distribution, not hardware, is the gate on delivery robots — riding an existing ordering platform is how a fleet gets demand without building a consumer brand.
- Market & Business
Relativity Networks raises $22M for faster fiber inside data centers
Interconnect is the quiet ceiling on cluster scale — capital moving into fiber is a read on where the bottleneck is now, not where the GPUs are.
- AI in Education✓ verified
ChatGPT for Teens ships globally with an age-prediction model deciding who gets it
Anyone building a learning product now has a reference implementation of the compliance shape regulators are pushing toward: infer the age, degrade the experience automatically, and make the pedagogy (guided questions, not answers) the safety feature rather than a bolt-on filter.
- Open Source & Self-Hostable
Ollama halves time-to-first-token by caching resolved model metadata
Time-to-first-token is what a local model feels like, so halving it changes whether self-hosting is pleasant enough to use for real work.
- Market & Business✓ verified
Jasper grounds its answer-engine-optimization agent in a brand’s own Search Console data
Grounding the recommendation in first-party Search Console data is the difference between an AEO tool you can measure and one that just sounds confident — ask for that grounding before buying any of them.
- AI in Education✓ verified
Google puts study tools into Search and a learning hub into Gemini
The free tier of a general assistant now covers most of what a standalone study app sold last year — flashcards, quizzes, notebooks, a photo-to-explanation loop — which resets the floor any paid learning product has to clear to justify its price.