Topic
Curating what information enters an LLM's context window at inference time, including retrieval, compression, memory and tool-result filtering.
Google has moved its Docsy documentation theme to the Linux Foundation and added experimental features that serve pages to AI agents. For teams on the theme, feeding docs to agents is now a setting, so the work moves to keeping the pages agents act on current.
Reality
Anthropic's open SKILL.md format is read by roughly forty agent tools, among them Codex, Cursor, Copilot and Gemini CLI. Each installed skill costs about 100 tokens a session until a task matches its description and the full workflow loads.
Reality
Platinum River Innovations told theCUBE it moved from time-and-materials to fixed bids and took the delivery risk onto itself. The wider repricing claim made at the Certinia event rests on that one named firm.
Reality
With documentation and web search removed, GPT-5.6 Luna passed 18% of 336 Dev Proxy tasks and 15% of 413 SPFx tasks, and the passes appear throughout both product histories instead of stopping at one release.
Reality
Dhravya Shah spent three days probing the personal agent from the outside and reports git-tracked Markdown searched by keyword, with a background pass he clocked taking 23 hours and 16 minutes to commit one preference.
Reality
Justin Johnsen has been a forward-deployed engineer at KPMG for about a year. His account of one engagement puts months of requirements and context work ahead of a build that took roughly a month, then coaching the client's own team.
Reality
A dev.to piece on harness engineering splits agent degradation into five separable failure modes. The two it actually describes both turn on where a token sits in the assembled prompt, and a bigger window leaves both in place.
Reality
In a dev.to walkthrough, miruky uses one bad rstrip call to separate five layers of agent engineering. Which layer do you change when a repair passes all three examples and still accepts 1e3s?
Reality
shinpr has taken claude-code-workflows through 133 releases, and the recent ones delete structure the models no longer need. His session reader then found three mandatory steps in his own repository that never ran.
Reality
Ajay Prakash's InfoQ talk walks a pager alert through logs, a downstream service and a buggy pull request. Each hop in that path depends on debugging instructions another team at LinkedIn wrote down.
Reality
A field-notes post on production LLM agents argues the context window behaves like a cache under eviction pressure, and that the real work is deciding every turn what earns space and which failed turns to scrub.
Reality
Son Nguyen says his team ran the same prompt and the same use case through different models, found the difference surprisingly small, and rebuilt the preprocessing, context and validation around the model instead. He calls that surrounding system the agent harness.
Reality
Newer models ship with advice to keep the guidance file under 200 lines. The analyst who wrote those 1,042 lines argues each one logs context the model lacked, and that the real debt is a mechanical rule left sitting in prose.
Reality
The harness lost its hidden system prompt, 43% of its builtin tool descriptions and its todo list middleware. LangChain's own footnote says reward confidence intervals span zero for every model tested, so the evals settle the token saving more firmly than the quality.
Reality
Polylane moved triage, investigation and code generation into a single run on September 3. Model spend per pull request fell from $111 to about $18, though the agent now files seven times as many of them.
Reality
Durable state and sandboxed execution now arrive as platform primitives, so a dev.to guide argues the remaining failure mode is undocumented organizational context. It leans on one arXiv paper and one example.
Reality
Alberto Souza says he reviewed none of the generated code. The checking he does do happens before any code exists, in context files and successive rounds of questions from the agent that he answers one by one.
Reality
Writing on dev.to, Ben McCarthy describes rebuilding the guest-messaging agents across his 23 holiday lets around a typed per-request view of state, with world facts held apart from the policy about what the agent may promise.
Reality
Anton Brilliantov spent eight parts specifying agent handoffs down to a single acceptance command. His ninth names the task shapes where those facts are still unknown, and the reading that has to happen first.
Reality
A dev.to post hands four models a file whose comment contradicts its code, then asks a clean session to fix the inconsistency. It is a well-built way to expose the failure, and no counts are published yet.
Reality
Product · August 19, 2026 · Confirmed2 publishers