DEVELOPER WORKFLOW

A July 2026 paper from Tencent Hunyuan turns agent harnesses into behavior-level maps. The useful lesson for builders is simple: code search is not enough when one behavior spans prompts, tools, state, permissions, and runtime policy.

SkillHone is a July 2026 paper about evolving agent skills across sessions. The useful takeaway for developers is simple: do not save only the latest SKILL.md. Save the decisions that explain why it changed.

Long-Horizon-Terminal-Bench tests coding agents on 46 terminal tasks that can run for 90 minutes. The takeaway is not that agents are useless. It is that evals need to measure endurance, recovery, and partial progress.

A Microsoft field study found that CLI coding-agent adoption spreads through peers and managers, while adopters merged roughly 24% more pull requests. The lesson is not to buy more seats. It is to instrument rollout, retention, cost, and review quality from day one.

ByteDance's Dockerless paper asks whether coding-agent patches can be verified without spinning up per-repo environments. The practical answer is not replace CI. It is use cheaper evidence before CI.

A new Vera paper tests Codex, Claude Code, OpenClaw, and Hermes with executable safety cases. The useful lesson is not panic. It is evidence-grounded agent QA.

The Program-as-Weights paper is a useful signal for developers: some LLM calls may move from per-request API prompts into compact local artifacts that behave like reusable fuzzy functions.

OpenAI's workplace agent data points to a practical shift: non-developers are starting to use agents for real work, so engineering teams need paved paths, policy, and receipts.

GitHub's June Copilot review updates point to a practical policy stack for agent-authored pull requests: validation, review depth, repo instructions, attribution, and release-note accountability.

AI agents are getting their own computers. Here is how to choose a sandbox architecture: filesystem isolation, network policy, secrets boundaries, snapshots, and when shell access is overkill.

Aharness, LangChain's custom harness pattern, and OpenAI's code-first migration all point to the same next step: agent processes need typed gates, validated evidence, and controlled transitions.

The Bayer and Thoughtworks PRINCE case study is a useful reminder that reliable agentic AI comes from context routing, traces, evals, monitoring, and human review, not from a better prompt alone.

Goal, loop, routine. Three verbs, two tools, one hard part. A complete field guide to running agentic loops in Claude Code and Codex, the real commands, the patterns people actually run, and the two failure modes that burn money.

MCP's new enterprise-managed authorization flow is not just less login friction. It moves agent tool access into identity, policy, and audit systems enterprises already understand.

Cohere shipped its first developer-facing model on June 9, 2026. North Mini Code is a 30B mixture-of-experts coding model with 3B active parameters, Apache 2.0 weights, and a deployment footprint of a single H100. Here is what it actually offers and where the open questions are.

The viral DN42 AWS bill story is funny until you realize the missing primitive: infrastructure agents need hard cloud-spend guardrails before they touch real accounts.

Choosing a local coding LLM in 2026 means balancing benchmark performance, hardware cost, and the compliance pressure to keep code off third-party servers. Here is what to run and on what hardware.

A Hacker News thread on config files that run code points at the next AI coding risk: agent hooks, skills, and editor rules need review like executable dependencies.

OpenAI's harness engineering post and new token-use research point to the same lesson: agentic coding teams need token budgets, receipts, and eval loops, not vibes.

The rsync Claude debate shows why teams need reproducible defect forensics before AI attribution becomes a public blame machine.