--- title: Agentic Coding description: "Agentic coding is software development where AI agents do more than autocomplete code: they plan, edit, test, review, open PRs, respond to feedback, and sometimes merge or deploy under human supervision. The imported corpus treats it as a shift from hand-authoring every line to designing the harness, context, constrain." created: 2026-05-22 updated: 2026-07-08 timestamp: "2026-07-08T00:00:00Z" type: concept tags: [agents, software, automation, evaluation, context] sources: [raw/imports/knowledge/agentic-coding/2026-03-05-harness-engineering-openai-codex.md, raw/imports/knowledge/agentic-coding/2026-03-10-cicd-context-agentic-coding-tessl.md, raw/imports/knowledge/agentic-coding/2026-03-17-claude-code-architecture-handbook.md, raw/imports/knowledge/references/pragmatic-engineer/revisiting--no-silver-bullets--in-the-age-of-ai.md, raw/articles/blogwatcher/2026-05-24/2026-05-23-latent-space-ainews-all-model-labs-are-now-agent-labs.md, raw/articles/blogwatcher/2026-05-24/2026-05-22-simon-willison-the-memory-shortage-is-causing-a-repricing-of-consumer-electronics.md, raw/imports/knowledge/research/tessl/2026-05-06-hack-night-london-tessl.md, raw/imports/knowledge/research/tessl/2026-05-06-evaluating-ai-real-world.md, raw/articles/blogwatcher/2026-05-24/2026-05-22-tessl-fireside-chat-guy-podjarny.md, raw/articles/blogwatcher/2026-05-24/2026-05-22-tessl-ai-native-devcon-ldn.md, raw/articles/blogwatcher/2026-05-25/2026-05-24-simon-willison-datasette-agent-0-1a4.md, raw/articles/blogwatcher/2026-05-25/2026-05-24-simon-willison-quoting-armin-ronacher.md, raw/imports/knowledge/research/tessl/2026-05-20-ai-native-devcon26.md, raw/imports/knowledge/research/tessl/2026-05-20-coding-agent-failure-patterns.md, raw/articles/blogwatcher/2026-05-25/2026-05-01-tessl-from-blind-spots-to-merged-prs.md, raw/imports/knowledge/ai-agents/latent-space/new-ai-infra-unicorns-exa-modal-turbopuffer.md, raw/imports/knowledge/ai-agents/latent-space/giving-agents-computers-ivan-burazin-daytona.md, raw/imports/knowledge/ai-agents/latent-space/2026-05-20-railway.md, raw/imports/knowledge/references/pragmatic-engineer/the-pulse-antigravity-20-takes-ide-out-of-its-new-ide.md, raw/imports/knowledge/ai-agents/simon-willison/datasette-agent.md, raw/imports/knowledge/ai-agents/simon-willison/datasette-agent-sprites-0.1a0.md, raw/imports/knowledge/ai-agents/simon-willison/datasette-agent-charts-0.1a2.md, raw/imports/knowledge/ai-agents/simon-willison/datasette-agent-0.1a3.md, raw/imports/knowledge/ai-agents/simon-willison/2026-05-20-google-io.md, raw/imports/knowledge/ai-agents/latent-space/2026-05-20-ainews-google-io-2026-gemini-35-flash-omni-spark-antigravity-20.md, raw/imports/knowledge/references/pragmatic-engineer/2026-05-19-ai-impact-on-software-engineers-part-2.md, raw/articles/blogwatcher/2026-05-28/2026-05-27-tessl-dont-make-your-agent-guess.md, raw/articles/blogwatcher/2026-05-28/2026-05-26-tessl-the-reinvention-of-the-dev-team.md, raw/articles/blogwatcher/2026-05-28/2026-05-26-interconnects-ai-some-ideas-for-what-comes-next-may.md, raw/articles/blogwatcher/2026-05-28/2026-05-27-latent-space-ainews-new-ai-infra-decacorns-fireworks.md, raw/articles/blogwatcher/2026-05-28/2026-05-27-pragmatic-engineer-building-opencode-with-dax-raad.md, raw/articles/blogwatcher/2026-05-28/2026-05-26-pragmatic-engineer-state-of-software-engineering-job-market-2026.md, raw/articles/blogwatcher/2026-05-28/2026-05-27-simon-willison-sqlite-agents-md.md, raw/articles/blogwatcher/2026-05-28/2026-05-27-simon-willison-openai-anthropic-product-market-fit.md, raw/articles/blogwatcher/2026-05-28/2026-04-30-tessl-warp-goes-open-source.md, raw/articles/blogwatcher/2026-05-28/2026-04-29-tessl-openai-open-sources-symphony.md, raw/articles/blogwatcher/2026-05-31/2026-05-30-simon-willison-how-we-contain-claude-across-products.md, raw/articles/blogwatcher/2026-05-31/2026-05-30-simon-willison-running-python-asgi-apps-in-the-browser-via-pyodide-a-service-worker.md, raw/articles/blogwatcher/2026-05-31/2026-05-30-latent-space-ainews-founders-and-forward-deployed-engineers.md, raw/articles/blogwatcher/2026-05-31/2026-05-12-pragmatic-engineer-revisiting-no-silver-bullets-in-the-age-of-ai.md, raw/articles/blogwatcher/2026-06-01/2026-05-31-simon-willison-the-solution-might-be-cancelling-my-ai-subscription.md, raw/articles/blogwatcher/2026-06-01/2026-05-22-simon-willison-pydantic-monty-investigation.md, raw/articles/blogwatcher/2026-06-01/2026-05-05-latent-space-doing-vibe-physics-alex-lupsasca-openai.md, raw/articles/blogwatcher/2026-06-01/2026-05-05-pragmatic-engineer-designing-data-intensive-applications-the-cloud-doing-the-right-thing.md, raw/articles/blogwatcher/2026-06-01/2026-05-04-latent-space-ainews-the-other-vs-the-utility.md, raw/articles/blogwatcher/2026-06-01/2026-05-02-latent-space-ainews-ai-engineer-world-s-fair-autoresearch-memory-world-models-tokenmaxxing-agentic-commer.md, raw/articles/blogwatcher/2026-06-01/2026-04-30-pragmatic-engineer-the-pulse-ai-load-breaks-github-why-not-other-vendors.md, raw/articles/blogwatcher/2026-06-01/2026-04-29-pragmatic-engineer-building-pi-and-what-makes-self-modifying-software-so-fascinating.md, raw/articles/blogwatcher/2026-06-01/2026-04-28-pragmatic-engineer-how-will-ai-change-operating-systems-part-1-ubuntu-and-linux.md, raw/articles/blogwatcher/2026-06-15/2026-06-14-simon-willison-why-ai-hasnt-replaced-software-engineers-and-wont.md, raw/articles/blogwatcher/2026-06-15/2026-06-05-latent-space-how-to-stop-shipping-low-quality-rl-environments-with-examples.md, raw/articles/blogwatcher/2026-06-15/2026-06-04-latent-space-reality-the-final-eval-lukas-petersson-and-axel-backlund-of-andon-labs.md, raw/articles/blogwatcher/2026-06-15/2026-06-03-latent-space-scaling-past-informal-ai-carina-hong-axiom-math.md, raw/articles/blogwatcher/2026-06-15/2026-06-03-pragmatic-engineer-kubernetes-and-retiring-at-the-top-with-kelsey-hightower.md, raw/articles/blogwatcher/2026-06-15/2026-06-03-simon-willison-uber-caps-usage-of-ai-tools-like-claude-code-to-manage-costs.md, raw/articles/blogwatcher/2026-06-15/2026-06-03-tessl-ai-native-devcon-day-2-from-agent-demos-to-operating-models.md, raw/articles/blogwatcher/2026-06-15/2026-06-06-simon-willison-running-python-code-in-a-sandbox-with-micropython-and-wasm.md, ../knowledge/raw/articles/blogwatcher/2026-06-26/2026-06-25-latent-space-ainews-its-meta-harness-summer.md, ../knowledge/raw/articles/blogwatcher/2026-06-26/2026-06-24-latent-space-why-the-frontier-ecosystem-must-be-open-matei-zaharia-and-reynold-xin-databricks.md, ../knowledge/raw/articles/blogwatcher/2026-06-26/2026-06-24-tessl-cursors-new-leaderboard-shows-teams-the-most-popular-plugins-skills-and-mcps.md, ../knowledge/raw/articles/blogwatcher/2026-06-26/2026-06-24-latent-space-ainews-claude-tag-multiplayer-proactive-persistent-agents-in-slack.md, ../knowledge/raw/articles/blogwatcher/2026-06-26/2026-06-23-tessl-the-new-tessl-review-now-you-decide-what-good-looks-like.md, ../knowledge/raw/articles/blogwatcher/2026-06-26/2026-06-23-tessl-ai-agent-governance-10-takeaways-from-engineering-leaders-on-agentic-development.md, ../knowledge/raw/articles/blogwatcher/2026-06-26/2026-06-23-tessl-why-warp-is-betting-engineering-leaders-are-done-picking-a-favourite-coding-agent.md, ../knowledge/raw/articles/blogwatcher/2026-06-26/2026-06-23-pragmatic-engineer-slow-down-to-speed-up-so-much-has-changed-in-6-months-time.md, ../knowledge/raw/articles/blogwatcher/2026-06-26/2026-06-24-pragmatic-engineer-tech-interviews-with-neetcode.md, ../knowledge/raw/articles/blogwatcher/2026-06-26/2026-06-22-simon-willison-porting-the-moebius-0-2b-image-inpainting-model-to-run-in-the-browser-with-claude-code.md] confidence: medium contested: false contradictions: [] --- # Agentic Coding ## Definition Agentic coding is software development where AI agents do more than autocomplete code: they plan, edit, test, review, open PRs, respond to feedback, and sometimes merge or deploy under human supervision. The imported corpus treats it as a shift from hand-authoring every line to designing the harness, context, constraints, and verification loops that let agents work safely. ## Current Synthesis Several sources converge on the same design lesson: agent output improves when the repository is legible to agents. Instructions should be versioned, discoverable, and mechanically checked. AGENTS.md or CLAUDE.md should act as a table of contents or collaboration contract rather than a huge prompt dump. Deeper docs, lint rules, tests, screenshots, observability, and issue templates become part of the agent runtime. OpenAI's Codex harness-engineering source describes an agent-first experiment where humans steer and agents execute, supported by structured docs, strict architectural layers, worktree-local environments, observability, and background cleanup tasks. Tessl's context CI/CD source argues that agent context needs its own eval and deployment discipline because context staleness and instruction changes can silently alter behavior. The Claude Code architecture source adds a six-layer model: context, tools, [[agent-skills]], hooks, subagents, and verification. ## Tensions - The Pragmatic Engineer source cautions that AI code generation has not yet proven a true "silver bullet" for inherent software complexity. - Agent-first sources report high throughput, but the durability of productivity gains depends on verification, architecture, and maintenance rather than raw code volume. - More instructions can create whack-a-mole effects; adding one rule shifts overall behavior rather than acting like a deterministic program. ## Related Pages - [[claude-code]] - [[openclaw]] - [[agent-context-management]] - [[agent-skills]] ## Blogwatcher Ingest Notes ### 2026-05-24 New blogwatcher sources reinforce the shift from model-only progress to production agent systems. Latent Space framed model labs as increasingly becoming agent labs; Tessl event pages centered evaluation, skill supply chains, context, guardrails, and AI-native development; Simon Willison's memory-shortage link notes AI datacenter demand spilling into hardware economics. Together these support the page's existing claim that durable gains come from harnesses, evaluation, workflow design, and infrastructure rather than raw code generation alone. Related: [[agent-skills]], [[agent-context-management]], [[openclaw]]. ### 2026-05-26 This batch reinforces that agentic coding is broadening from IDE assistants into agent-native clouds, terminals, application plugins, and workflow surfaces. Latent Space items on Daytona/Railway and Google I/O, Pragmatic Engineer coverage of Antigravity and software-engineering impacts, and Simon Willison's Datasette Agent posts all point to agents being given more complete computers, deployment environments, and product UI entry points rather than just code-completion slots. Related: [[agent-context-management]], [[agent-evaluation]], [[openclaw]]. ### 2026-05-28 This batch strengthens the view that coding agents are becoming an economic and organizational force, not just an IDE feature. Tessl and Pragmatic Engineer sources emphasize that implementation speed shifts the bottleneck toward product judgment, task specification, verification, ownership, and team design; Simon Willison argues Claude Code/Codex-style products have created enterprise product-market fit through high token-consuming daily workflows; OpenCode and Warp show open-source/positioning/community strategies around coding harnesses; and SQLite's AGENTS.md shows maintainers setting boundaries for agentic contributions. Related: [[agent-evaluation]], [[agent-security]], [[agent-context-management]], [[openclaw]]. ### 2026-05-29 The newest high-signal sources push agentic coding toward async/background work: Cognition/Devin is framed around agents with repo access, machines, memory, tests, and review loops working away from the developer's main flow; Simon Willison's LLM talk and Datasette Agent notes show agent loops being embedded into tools and products; and Pragmatic Engineer's AI-spend item suggests token budgets and agent usage are becoming engineering-management concerns. Related: [[agent-context-management]], [[agent-evaluation]], [[openclaw]]. ^[raw/articles/blogwatcher/2026-05-29/2026-05-28-latent-space-the-age-of-async-agents-cognition-s-walden-yan-openinspect-s-cole-murray.md] ^[raw/articles/blogwatcher/2026-05-29/2026-05-19-simon-willison-the-last-six-months-in-llms-in-five-minutes.md] ^[raw/articles/blogwatcher/2026-05-29/2026-05-28-pragmatic-engineer-the-pulse-a-trend-of-trying-to-cut-back-on-ai-spend-within-eng-departments.md] ### 2026-05-30 This batch adds three durable agentic-coding signals: Claude Opus 4.8 is framed as an incremental but useful long-horizon coding-agent update; Claude Code Dynamic Workflows/“ultracode” point toward massively parallel subagent orchestration; and Datasette/Pragmatic Engineer items show agent-assisted development and forward-deployed engineering becoming mainstream software-workflow concerns. Related: [[agent-evaluation]], [[agent-context-management]], [[claude-code]]. ^[raw/articles/blogwatcher/2026-05-30/2026-05-29-tessl-ai-coding-agent-accuracy-opus-4-7-vs-4-8.md] ### 2026-05-31 This batch adds more evidence that agentic coding is blending with forward-deployed engineering, local/browser runtimes, coding-agent containment, and normal software-delivery economics. The most durable items were the Claude containment note, the Pyodide/ASGI browser experiment, Pragmatic Engineer's AI/no-silver-bullets framing, and Latent Space's forward-deployed-engineer coverage. Related: [[agent-security]], [[datasette]], [[anthropic]], and [[agent-evaluation]]. ^[raw/articles/blogwatcher/2026-05-31/2026-05-30-simon-willison-how-we-contain-claude-across-products.md] ^[raw/articles/blogwatcher/2026-05-31/2026-05-30-simon-willison-running-python-asgi-apps-in-the-browser-via-pyodide-a-service-worker.md] ^[raw/articles/blogwatcher/2026-05-31/2026-05-30-latent-space-ainews-founders-and-forward-deployed-engineers.md] ^[raw/articles/blogwatcher/2026-05-31/2026-05-12-pragmatic-engineer-revisiting-no-silver-bullets-in-the-age-of-ai.md] ### 2026-06-01 This batch reinforces the social and systems side of AI-assisted development: Simon Willison's subscription-cancellation note frames overuse/misuse risk for coding agents; pydantic-monty preserves a concrete Python/package-investigation workflow; Pragmatic Engineer items connect AI pressure to GitHub load, self-modifying software, and operating-system direction; and Latent Space keeps agentic commerce, memory, world models, and research automation in view. Related: [[agent-security]], [[agent-evaluation]], [[datasette]], and [[anthropic]]. ^[raw/articles/blogwatcher/2026-06-01/2026-05-31-simon-willison-the-solution-might-be-cancelling-my-ai-subscription.md] ^[raw/articles/blogwatcher/2026-06-01/2026-05-22-simon-willison-pydantic-monty-investigation.md] ^[raw/articles/blogwatcher/2026-06-01/2026-05-05-latent-space-doing-vibe-physics-alex-lupsasca-openai.md] ^[raw/articles/blogwatcher/2026-06-01/2026-05-05-pragmatic-engineer-designing-data-intensive-applications-the-cloud-doing-the-right-thing.md] ### 2026-06-13 This Blogwatcher batch adds a fresh cluster around coding-agent economics and governance: model billing/name mismatches, context-engineering cost, reviewability as the bottleneck, agent skills/harnesses, enterprise operating models, Claude Fable behavior, Claude Code misuse/security reporting, and Simon Willison's concrete Datasette/MicroPython/WASM sandboxing examples. Related: [[agent-context-management]], [[agent-evaluation]], [[agent-security]], [[anthropic]], and [[datasette]]. ^[raw/articles/blogwatcher/2026-06-13/2026-06-12-simon-willison-quoting-andrew-singleton.md] ^[raw/articles/blogwatcher/2026-06-13/2026-06-12-tessl-why-your-gemini-bill-doesn-t-match-the-model-names.md] ^[raw/articles/blogwatcher/2026-06-13/2026-06-12-tessl-claude-fable-5-vs-opus-4-8-the-mythos-hype-meets-reality.md] ^[raw/articles/blogwatcher/2026-06-13/2026-06-12-latent-space-ainews-loopcraft-the-art-of-stacking-loops.md] ^[raw/articles/blogwatcher/2026-06-13/2026-06-11-simon-willison-claude-fable-is-relentlessly-proactive.md] ### 2026-06-14 This batch reinforces that agentic coding bottlenecks are moving from raw generation toward context engineering, reviewability, benchmark quality, and cost/model-accounting. Tessl, Latent Space, Pragmatic Engineer, and Simon Willison items all point back to the same operating model: better harnesses, clearer context, and stronger review loops around coding agents. Related: [[agent-context-management]], [[agent-evaluation]], [[agent-security]], and [[datasette]]. ^[raw/articles/blogwatcher/2026-06-14/2026-06-13-latent-space-ainews-fable-and-mythos-officially-too-dangerous-to-release.md] ^[raw/articles/blogwatcher/2026-06-14/2026-06-13-simon-willison-statement-on-the-us-government-directive-to-suspend-access-to-fable-5.md] ^[raw/articles/blogwatcher/2026-06-14/2026-06-12-simon-willison-openai-webrtc-audio-session-now-with-document-context.md] ^[raw/articles/blogwatcher/2026-06-14/2026-06-10-latent-space-ainews-anthropic-claude-fable-5-mythos-but-safe-with-controversial-ter.md] ### 2026-06-15 This batch adds another governance-and-reviewability cluster: Simon Willison and Tessl argue that AI has not removed software-engineering bottlenecks but moved them toward operating models, review, evals, and costs; Latent Space items on RL environments, reality evals, and formal/informal AI reinforce the need for better tasks and verification; and Pyodide/MicroPython examples show concrete sandboxed execution surfaces. Related: [[agent-evaluation]], [[agent-security]], [[datasette]], and [[claude-code]]. ^[raw/articles/blogwatcher/2026-06-15/2026-06-14-simon-willison-why-ai-hasnt-replaced-software-engineers-and-wont.md] ^[raw/articles/blogwatcher/2026-06-15/2026-06-05-latent-space-how-to-stop-shipping-low-quality-rl-environments-with-examples.md] ^[raw/articles/blogwatcher/2026-06-15/2026-06-04-latent-space-reality-the-final-eval-lukas-petersson-and-axel-backlund-of-andon-labs.md] ^[raw/articles/blogwatcher/2026-06-15/2026-06-03-latent-space-scaling-past-informal-ai-carina-hong-axiom-math.md] ### 2026-06-16 This batch reinforces the reviewability/operating-model thesis: Tessl DevCon/reviewability items, GitHub/Latent Space agent discussions, and Simon Willison Datasette-agent notes all frame production coding agents as systems requiring clear tasks, review loops, sandboxing, and cost/security controls. Related: [[agent-evaluation]], [[agent-security]], [[agent-context-management]]. ^[raw/articles/blogwatcher/2026-06-16/2026-06-15-simon-willison-datasette-agent-0-3a0.md] ^[raw/articles/blogwatcher/2026-06-16/2026-06-02-simon-willison-datasette-agent-micropython-0-1a0.md] ^[raw/articles/blogwatcher/2026-06-16/2026-06-02-latent-space-github-s-plan-for-agents-kyle-daigle-github.md] ^[raw/articles/blogwatcher/2026-06-16/2026-06-02-pragmatic-engineer-ideas-slow-down-to-speed-up-when-working-with-ai-agents.md] ### 2026-06-17 This Blogwatcher batch adds a fresh cluster around coding-agent economics and governance: model billing/name mismatches, context-engineering cost, reviewability as the bottleneck, agent skills/harnesses, enterprise operating models, Claude Fable behavior, Claude Code misuse/security reporting, and Simon Willison's concrete Datasette/MicroPython/WASM sandboxing examples. Related: [[agent-context-management]], [[agent-evaluation]], [[agent-security]], [[anthropic]], and [[datasette]]. ^[raw/articles/blogwatcher/2026-06-17/2026-06-16-simon-willison-datasette-tailscale-0-1a0.md] ^[raw/articles/blogwatcher/2026-06-17/2026-06-16-simon-willison-quoting-georgi-gerganov.md] ^[raw/articles/blogwatcher/2026-06-17/2026-06-16-interconnects-ai-frontier-post-training-recipe-review-with-finbarr-timbers.md] ^[raw/articles/blogwatcher/2026-06-17/2026-06-16-simon-willison-the-fable-5-export-controls-harm-us-cyber-defense.md] ^[raw/articles/blogwatcher/2026-06-17/2026-06-16-simon-willison-quoting-matteo-wong-the-atlantic.md] ### 2026-06-19 This batch sharpens the open-vs-frontier coding-agent thesis: Tessl's ~1,000-scenario eval has GLM 5.2 beating Claude Sonnet 4.6 on both quality and cost, and GitHub Copilot CLI's "Rubber Duck" (a second model, GPT-5.4, reviewing a Claude agent) closed 74.7% of the Sonnet→Opus gap on SWE-Bench Pro — cross-model review as a verification layer. Linear's "issue tracking is dead" frames raw context (not tickets) as the unit of work for agents; Charity Majors' "code became disposable overnight" and the "bright kid with no manners" evals-can-lie piece stress that cheaper code demands more discipline, rules, and scripts. CI/CD priorities shift from speed to risk once agents babysit slow pipelines. Related: [[glm]], [[agent-evaluation]], [[agent-security]]. ^[raw/articles/blogwatcher/2026-06-19/2026-06-18-tessl-open-source-coding-agents.md] ^[raw/articles/blogwatcher/2026-06-19/2026-04-08-tessl-github-rubber-duck.md] ^[raw/articles/blogwatcher/2026-06-19/2026-03-31-tessl-linear-agents.md] ^[raw/articles/blogwatcher/2026-06-19/2026-03-26-tessl-bright-kid-no-manners-part-2.md] ^[raw/articles/blogwatcher/2026-06-19/2026-06-17-simon-willison-charity-majors.md] ^[raw/articles/blogwatcher/2026-06-19/2026-06-17-pragmatic-engineer-cicd-robert-erez.md] ### 2026-06-20 This batch adds two durable agentic-coding threads. (1) **Tessl "bright kid, part 1"** isolates the binding constraint as **social/process context, not code correctness**: 16 failure modes (unread CONTRIBUTING.md, forged DCO sign-off, sniped claimed issues, unsolicited refactoring of already-rejected approaches), all fixed by a three-layer tile (skills/rules/scripts) that takes the same agent from 15% → 99% process-compliance with unchanged code. It is a concrete case of the "harness/context beats raw model" thesis already tracked here. (2) **Latent Space AINews** flags a workflow-scale problem: traditional git/GitHub workflows **break under dozens-to-hundreds of concurrent code agents** (stale worktrees, diverged review state, env-setup overhead, poor state sync). @_xjdr proposes **Noumena Code / ncode** — virtual shallow checkouts + jj + Sapling-style commit stacks + cloud sync + file-level ACLs + vertical integration from model to SCM to remote runtimes. Adjacent releases: OpenAI Codex Record & Replay (workflow → inspectable skill), Cursor /automate (NL → configured triggers/tools), Claude Code Artifacts, Devin Review/Security. Related: [[glm]], [[agent-evaluation]], [[agent-security]], [[agent-skills]]. ^[raw/articles/blogwatcher/2026-06-20/2026-03-26-tessl-bright-kid-no-manners-part-1.md] ^[raw/articles/blogwatcher/2026-06-20/2026-06-19-latent-space-ainews-glm-52-vibe.md] ### 2026-06-26 This batch makes the **meta-harness/control-plane** layer explicit. Latent Space and Databricks frame Omnigent as an open, pluggable layer above Claude Code, Codex, Cursor, Pi and internal agents; Cursor 3.9 adds a team/workspace/user control plane for plugins, skills, MCP servers, subagents, rules, commands and hooks; Warp/Oz argues teams will route work across multiple harnesses/models with shared memory, budgets and ROI metrics; and Claude Tag moves Claude Code into Slack as a multiplayer async/proactive coworker. Tessl and Pragmatic Engineer add the operating-model caution: agentic coding scales only when teams define what “good” looks like, govern rollout/cost/identity/permissions, and preserve human review for higher-risk work. Simon Willison’s Cloudflare temporary-deployment note adds a concrete deployment primitive for agent-built experiments: `wrangler deploy --temporary` creates a 60-minute Worker project without an account, useful as a bounded throwaway deployment surface. Related: [[agent-context-management]], [[agent-evaluation]], [[agent-security]]. ^[../knowledge/raw/articles/blogwatcher/2026-06-26/2026-06-25-latent-space-ainews-its-meta-harness-summer.md] ^[../knowledge/raw/articles/blogwatcher/2026-06-26/2026-06-24-latent-space-why-the-frontier-ecosystem-must-be-open-matei-zaharia-and-reynold-xin-databricks.md] ^[../knowledge/raw/articles/blogwatcher/2026-06-26/2026-06-24-tessl-cursors-new-leaderboard-shows-teams-the-most-popular-plugins-skills-and-mcps.md] ^[../knowledge/raw/articles/blogwatcher/2026-06-26/2026-06-23-tessl-why-warp-is-betting-engineering-leaders-are-done-picking-a-favourite-coding-agent.md] ^[../knowledge/raw/articles/blogwatcher/2026-06-26/2026-06-21-simon-willison-temporary-cloudflare-accounts-for-ai-agents.md] ### 2026-06-27 This batch reinforces that agentic coding is constrained by **model capability floors, delegation skill, and organizational adoption**, not simply by access to cheap tokens. Tessl's Nemotron test says 30B-class models can be economical for narrow tasks but fall below reliable multi-step agent behavior; Latent Space records OpenAI internal Codex token use rising sharply across Research, Support, Engineering, and Legal; Timothy B. Lee's quoted analogy frames LLM use as a management/delegation skill rather than a no-learning-curve tool. Related: [[agent-evaluation]], [[agent-security]], [[glm]]. ^[../knowledge/raw/articles/blogwatcher/2026-06-27/2026-06-26-tessl-how-small-can-an-agent-model-get-the-nemotron-floor.md] ^[../knowledge/raw/articles/blogwatcher/2026-06-27/2026-06-26-latent-space-ainews-openai-reports-median-internal-codex-output-tokens-grew.md] ^[../knowledge/raw/articles/blogwatcher/2026-06-27/2026-06-26-simon-willison-quoting-timothy-b-lee.md] ### 2026-06-28 This quiet two-item batch reinforces the same agentic-coding operating-model thesis from two sides. Tessl's AI Engineering World's Fair post frames the next layer as **agent enablement**: a Tessl Agent watches pull requests, session logs, and tickets for recurring mistakes, then turns patterns into improvement PRs, workflows, GitHub Actions, and a context-development feedback loop. Latent Space's GPT-5.6 recap adds the market/governance side: frontier coding models are now evaluated and deployed with access controls, cyber thresholds, cheating-adjusted evals, cost tiers, and latency claims attached. Related: [[agent-context-management]], [[agent-evaluation]], [[glm]]. ^[../knowledge/raw/articles/blogwatcher/2026-06-28/2026-06-27-tessl-see-you-at-ai-engineering-worlds-fair-2026.md] ^[../knowledge/raw/articles/blogwatcher/2026-06-28/2026-06-27-latent-space-ainews-openai-gpt-5-6-sol-terra-luna.md] ### 2026-06-29 This batch adds a concise human-authority framing to the existing reviewability thesis: Jon Udell's quote (via Simon Willison) rejects “human in the loop” as machine-centered language and reframes agentic development as a human/team loop that invites agents into reviewable work. Interconnects' Artifacts #22 keeps open models in the coding-agent substrate conversation, while OpenClaw `v2026.6.11-beta.2` shows the operator-control plane continuing to harden around channel controls, file-driven runs, wake bridges, model/provider resolution, and delivery/session reliability. Related: [[agent-evaluation]], [[agent-context-management]], [[openclaw]], [[glm]]. ^[../knowledge/raw/articles/blogwatcher/2026-06-29/2026-06-28-simon-willison-quoting-jon-udell.md] ^[../knowledge/raw/articles/blogwatcher/2026-06-29/2026-06-28-interconnects-ai-artifacts-22-zyphra-cohere-and-poolside.md] ^[../knowledge/raw/articles/blogwatcher/2026-06-29/2026-06-28-openclaw-github-openclaw-2026-6-11-beta-2.md] ### 2026-06-30 Simon Willison's Ornith-1.0 note adds another local/open coding-model capability signal: the interesting part is not only benchmark positioning, but that a 35B GGUF run via LM Studio could stay useful across many tool calls in an agent harness against a Datasette checkout. This reinforces the page's existing emphasis on harness behavior, local execution, and multi-step tool reliability as the practical test for agentic-coding models. Related: [[agent-evaluation]], [[glm]], [[datasette]]. ^[../knowledge/raw/articles/blogwatcher/2026-06-30/2026-06-29-simon-willison-ornith-1-0.md] ### 2026-07-01 This batch pushes the page's thesis from "agents write code" toward **loop and control-plane engineering**. Tessl Agent turns recurring agent mistakes into skills, GitHub Actions, and scheduled improvement loops; Latent Space's Loopcraft issue says the leverage frontier is designing loops that prompt agents; Pragmatic Engineer frames OpenAI/Anthropic/Cursor investment around cloud agents that run asynchronously in managed environments; and Simon Willison's `shot-scraper video` gives agents a concrete verification/demo primitive. Related: [[agent-context-management]], [[agent-evaluation]], [[openclaw]], [[anthropic]]. ^[../knowledge/raw/articles/blogwatcher/2026-07-01/2026-06-30-tessl-agents-keep-making-same-mistakes.md] ^[../knowledge/raw/articles/blogwatcher/2026-07-01/2026-06-12-latent-space-loopcraft-art-of-stacking-loops.md] ^[../knowledge/raw/articles/blogwatcher/2026-07-01/2026-06-30-pragmatic-engineer-impressions-openai-anthropic-cursor.md] ^[../knowledge/raw/articles/blogwatcher/2026-07-01/2026-06-30-simon-willison-shot-scraper-video.md] ### 2026-07-02 This batch makes **software factories** the dominant agentic-coding frame: Introspection's autoresearch loop, Cursor's enterprise FDE program, Warp/Oz's factory floor, Sierra's agent-engineering role, and AIEWF's Loopcraft dispatch all treat engineering leverage as designing the lifecycle that collects signals, triages work, specifies tasks, implements, reviews, verifies, ships, monitors, and feeds back into the next loop. Kent Beck's Pragmatic Engineer interview adds the balancing constraint: AI may accelerate code accumulation, but software engineering still depends on trust, human coordination, and phase-appropriate judgment. Related: [[agent-context-management]], [[agent-evaluation]], [[anthropic]]. ^[../knowledge/raw/articles/blogwatcher/2026-07-02/2026-07-01-latent-space-autoresearch-introspection.md] ^[../knowledge/raw/articles/blogwatcher/2026-07-02/2026-07-01-latent-space-cursor-forward-deployed-engineers.md] ^[../knowledge/raw/articles/blogwatcher/2026-07-02/2026-07-01-latent-space-warp-software-factories.md] ^[../knowledge/raw/articles/blogwatcher/2026-07-02/2026-07-01-latent-space-forward-deployed-engineers.md] ^[../knowledge/raw/articles/blogwatcher/2026-07-02/2026-07-01-pragmatic-engineer-kent-beck-shapes-software.md] ### 2026-07-03 This batch adds the counterweight to yesterday's software-factory framing: Litt's "understand to participate," Osmani's inner-loop/outer-loop split, and Bakaus's "no auto" design stance all argue that humans must retain enough understanding, taste, and responsibility to steer agent work instead of delegating comprehension away. Simon Willison's `llm-coding-agent` 0.1a0 is the concrete harness-side artifact: a small Claude Code-style agent built with Fable 5, explicit file/command/search tools, command allowlists, tests, docs, and a Python API. Related: [[agent-skills]], [[agent-evaluation]], [[datasette]], [[openclaw]]. ^[../knowledge/raw/articles/blogwatcher/2026-07-03/2026-07-02-simon-willison-understand-to-participate.md] ^[../knowledge/raw/articles/blogwatcher/2026-07-03/2026-07-02-latent-space-aiewf-daily-dispatch-agency.md] ^[../knowledge/raw/articles/blogwatcher/2026-07-03/2026-07-02-latent-space-skill-engineering-design.md] ^[../knowledge/raw/articles/blogwatcher/2026-07-03/2026-07-02-simon-willison-llm-coding-agent-0-1a0.md] ### 2026-07-04 This batch extends the software-factory thesis into governance and economics. Latent Space's loop debate separates deterministic control loops from agentic hype and frames developers as "locomotive engineers" keeping automated work on the rails; the AI Engineer Survey reports agents are widely used and increasingly write data, but controls remain primitive and costs constrain ambition. Vercel's Andrew Qu adds the platform pattern: agents need primitives for context, tools, resumability, sandboxes, long-running work, observability, evals, skills, and agent-readable web surfaces. Simon Willison's Fable note adds an orchestration tactic: delegate implementation to lower-power subagents while keeping judgment/review in the main loop. Related: [[agent-context-management]], [[agent-skills]], [[claude-code]], [[agent-evaluation]]. ^[../knowledge/raw/articles/blogwatcher/2026-07-04/2026-07-03-latent-space-aiewf-daily-dispatch-loops.md] ^[../knowledge/raw/articles/blogwatcher/2026-07-04/2026-07-03-latent-space-vercel-agents-new-software.md] ^[../knowledge/raw/articles/blogwatcher/2026-07-04/2026-07-03-simon-willison-fables-judgement.md] ### 2026-07-05 Simon Willison's link to Armin Ronacher's Pi report adds a useful counterexample to simple model-progress narratives: newer Claude models can be better inside Claude Code's native edit-tool regime while becoming worse at a similar third-party edit schema that rejects invented nested fields. The durable implication for agentic-coding harnesses is that tool schemas, model-specific training, and fallback edit mechanisms may now be part of portability engineering rather than neutral plumbing. Related: [[claude-code]], [[agent-evaluation]], [[agent-context-management]]. ^[../knowledge/raw/articles/blogwatcher/2026-07-05/2026-07-04-simon-willison-better-models-worse-tools.md] ### 2026-07-06 Simon Willison's `sqlite-utils 4.0rc2` report is a compact agentic-coding case study: Claude Fable performed a final release-blocker review and implemented substantial transaction-semantics fixes, GPT-5.5 then cross-reviewed the diff and found two additional P1 edge cases, and Willison used `agentsview` to estimate the unsubsidized session/subagent cost at $149.25. Together with OpenClaw `v2026.7.1-beta.2` adding `openclaw attach`, `on-exit` cron triggers, ClawRouter budget/model routing, Telegram Codex steering, scoped capability profiles, and Cursor Agent autoreview, the batch reinforces that production agentic coding is becoming a discipline of harness attachment, cross-model review, cost accounting, and operator-control surfaces rather than raw code generation alone. Related: [[datasette]], [[openclaw]], [[agent-evaluation]]. ^[../knowledge/raw/articles/blogwatcher/2026-07-06/2026-07-05-simon-willison-sqlite-utils-4-0rc2-mostly-written-by-claude-fable.md] ^[../knowledge/raw/articles/blogwatcher/2026-07-06/2026-07-05-openclaw-github-openclaw-2026-7-1-beta-2.md] ### 2026-07-07 `sqlite-utils 4.0rc3` reinforces the release-engineering side of [[agentic-coding]]: after rc2, Claude Fable 5 and GPT-5.5-assisted backlog work still uncovered enough substantive database-behavior changes to justify another release candidate before stable 4.0. The durable lesson is that coding agents can accelerate late-cycle issue/PR processing, but the human release loop still needs semantic judgment about breaking API changes, compatibility with upstream conventions, and whether a stable release is truly ready. Related: [[datasette]], [[agent-evaluation]]. ^[../knowledge/raw/articles/blogwatcher/2026-07-07/2026-07-06-simon-willison-sqlite-utils-4-0rc3.md] ### 2026-07-08 This batch turns the recent "agentic release engineering" thread into a broader harness thesis. Willison's `sqlite-utils 4.0` post is a concrete example of Fable/GPT-assisted late-cycle release work, where the agent found blockers by writing repro scripts but the human maintained semantic release judgment; Latent Space's Fable field guide argues for "unhobbling" new model classes, blindspot passes, implementation notes, and more demanding prompts; Lilian Weng's harness-engineering survey formalizes the harness as the deployable layer around models (workflow, tools, memory, permissions, subagents, evals); BAIR's data-systems essay says near-free intelligence shifts systems work toward data systems for/of/by agents; Pragmatic Engineer's hiring-market piece adds the social counterpressure — AI-generated resume/applicant noise is reducing trust and making networks/specific AI/FDE skillsets more valuable. Related: [[agent-context-management]], [[agent-evaluation]], [[datasette]]. ^[../knowledge/raw/articles/blogwatcher/2026-07-08/2026-07-07-simon-willison-sqlite-utils-4-0-now-with-database-schema-migrations.md] ^[../knowledge/raw/articles/blogwatcher/2026-07-08/2026-07-07-latent-space-ainews-the-field-guide-to-fable.md] ^[../knowledge/raw/articles/blogwatcher/2026-07-08/2026-07-04-lil-log-harness-engineering-for-self-improvement.md] ^[../knowledge/raw/articles/blogwatcher/2026-07-08/2026-07-07-bair-blog-intelligence-is-free-now-what-data-systems-for-of-and-b.md] ^[../knowledge/raw/articles/blogwatcher/2026-07-08/2026-07-07-pragmatic-engineer-tech-jobs-market-in-2026-part-3-hiring-managers-job-see.md]