Agent memory needs correction as well as storage Anthropic tests how Claude checks and reorganizes its own memories — Keep original evidence searchable, then test whether reorganized notes improve answers enough to justify their upkeep. The …
Cotra traces a possible AI-research feedback loop, but its one-year scenario depends on assumptions and mostly withheld evidence.
Hightouch nearly matches Opus by escalating uncertain decisions. Harder cases shrink the saving, and HN practitioners question the changing economics.
Replit explains the isolation behind its coding workspaces Replit explains how separate Linux machines now isolate coding workspaces — Builders can separate untrusted code, network permissions and credentials while keeping workspaces quick …
Warp’s Ben Holmes scores conversations, tool calls, and code changes to find recurring failures and propose reviewed repairs to agent instructions and setup.
Anthropic’s Lance Martin covers prompt debt, cache reuse, evaluation-driven optimization, and recoverable agents, with a detailed Q&A on overfitting and human judgment.
Neo4j and Blitzy use an interest-calculation example to explain traceable dependency graphs, approved plans, and old-versus-new behavioral tests.
Matt Wolfe edits an animated image, tests hosted video demos, and finds the practical limits behind speed claims, open weights, and free allowances.
AI can implement research ideas while missing which experiments matter. A short account of that gap, a reported tenfold speedup, and practical dissent from HN.
Former Cursor designer Ryo Lu uses a journal-app demo to show how editable plans, small changes, and collaboration help people develop their own judgment.
Hard Fork examines why chatbot use doesn't mean consent to data centers, job disruption, or AI imposed on schools.
A new hosted model, with downloads still promised Thinking Machines launches Inkling 2 Small, with downloads still pending — Developers gain another hosted option for coding and custom training, with independent tests showing different …
Google DeepMind’s Philipp Schmid combines a sandbox, skills, tools, proxy-managed credentials, hooks, and schedules in a finance-agent demonstration.
IBM’s Mixture of Experts separates targeted model repair from recursive self-improvement, then examines open weights and the limits of world-model simulations.
Sid Pednekar uses two hypothetical asset-management firms to explain problem selection, shared capabilities, and learning after AI deployment.
CentralAgent AI’s Sparsh Jain turns repeated corrections into cases, strategies, and skills maintained through a memory lifecycle and evaluation loops.
Coatue’s Frank Long breaks investment research into testable steps run by code, statistical models, language models, and people.
OpenAI's kernel-auditing agents sometimes exploited the wrong bug. Checking how an agent succeeds matters as much as checking its result.
Restricted access to a new security model OpenAI begins limited access to its new security model — Security teams gain a specialized option, but access restrictions, data handling and mixed test results matter before adopting it. GPT-6 …
Keycard and Temporal demonstrate an expense agent that preserves user delegation across services and resumes settlement after a worker crash.
Netflix’s Cameron Wolfe connects calibrated human review, automated judges, and production experiments to make subjective AI evaluations more reliable.
Matched coding tasks exposed silent scope changes and invented blockers. OpenAI measured the gap, but its training cause remains unresolved.
Midwest AI Summit speakers connect an AI charter, human accountability, maintenance costs, and deployed agents that check work orders and answer data questions.
AI-found Linux flaws still need human-checked fixes OpenAI reports 948 Linux flaws that let local users gain root — Reproducing a security failure helps prioritize repairs, but generated patches still need review and delivery through your …
IBM’s Martin Keen separates the need for semantic search from the need for a separate database, using a four-level retrieval pyramid.
Blackstone and Norm AI CTOs discuss usable data, replaceable workflows, durable expert tests, and when to build, prototype, or wait.
Brainbase’s Domino’s pricing case pairs explicit definitions, browser traces, and repeated tests to improve a narrowly scoped agent.
Wise’s Emil Chetty separates customer authorization from agent planning, with locked payment jobs and checks in the services that move money.
Cole Medin tests three coding-factory configurations, finds costly rework in long runs, and keeps open models on scoped implementation.
FlashProve ties checked tokens to a specific model, but its headline speed counts prompt processing and proving still costs 70–95 times ordinary inference.
A new downloadable coding model Ricursive releases a coding model with smaller-memory downloads — The release gives developers another model to test on their own hardware, with experimental compressed files and unusually detailed …
Alexander Shevchenko explains Ramp Labs’ work on financial retrieval, precomputed context and parallel security agents, with task-specific company results.
Nous Research’s Dillon Rolnick argues that agent platforms create switching costs, and makes the case for open foundations with replaceable models.
General Intuition’s Pim de Witte describes a pixels-to-actions model trained on labeled gameplay, with claims of transfer across robot controllers.
Hamel Husain compares six classifiers on leasing-assistant traces, then shows how to tune thresholds, check confidence scores and weigh cost.
AI can produce a result that passes its checks. Who can defend it, question its assumptions and repair it when something goes wrong?
Matt Wolfe builds reusable scripting and editing skills, then fixes a rough cut through feedback, waveform checks and saved workflow changes.
A shared patient brief, organized records, better questions and expert updates: Peter Yang's five-step caregiver workflow, with clinical decisions left to the care team.
A hosted model release, with downloads to follow Tencent launches Hy4 in hosted apps, with downloads still pending — The release adds image understanding and computer control, but local deployment still awaits downloadable files. Tencent …
Langfuse's banking case combines representative test data, calculator tools, separate quality gates and expert review to decide when an agent can ship.
Datasite's CTO explains how document edge cases changed its build-versus-buy decision, and how a private benchmark checks the downstream answers.
NatWest's routing funnel, human reviews and miscalibrated judges show what it takes to prepare a customer-facing banking agent for release.
Ben Stolovitz runs five or six coding agents, but finds human attention, engineering judgment, and the pleasure of making things harder to replace.
Basis co-founder Mitchell Troyanovsky explains how behavior specs, replayed decisions and lightweight simulations help teams trust production agents.
A local model identifies three small networks’ training tasks from their numerical weights. The failed fourth case and hour-long runs keep the claim grounded.
Checking which model answered Anthropic details a way to check which model answered — FlashProve makes model substitution more checkable, but its measured costs and narrow guarantees matter more than the announcement’s broad promise. The …
George Cameron breaks down an agent benchmark's $21 task cost, its over-$200 uncached estimate, and how to choose models against your own workload.
A complaint about an apparently automated support reply gets a warmer apology. Whether anyone fixed the bug remains unanswered.
AI-assisted discovery challenges a profession built around individual achievement. Three mathematicians argue for the joy of understanding; readers ask who sustains the people doing it.
Reading raw network parameters and writing Python to improve a model use different capabilities. Honnibal separates the experiments behind the headline.