# Dark Factory Dev > Research and analysis tracking autonomous software production: its systems, evidence, economics, risks, and terminology. The HTML pages are canonical. Markdown pages are synchronized representations generated from the same published source files. ## Start here - [The Factory](https://darkfactory.dev/factory): A living atlas of the systems required for autonomous software production. - [AI Glossary](https://darkfactory.dev/glossary): Evidence-linked definitions with relationships and public sources. - [In the News](https://darkfactory.dev/news): Ranked roundups linking to primary sources. - [Blog](https://darkfactory.dev/blog): Arguments and field notes. ## Machine-readable indexes - [Corpus Markdown sitemap](https://darkfactory.dev/sitemap.md) - [Corpus manifest](https://darkfactory.dev/agent.json) - [Site JSON Feed](https://darkfactory.dev/feed.json) - [Site Atom feed](https://darkfactory.dev/feed.xml) - [Glossary JSON](https://darkfactory.dev/glossary.json) - [Knowledge graph JSON](https://darkfactory.dev/factory/graph.json) ## Recent writing - [Generation is cheap. Verification is the factory.](https://darkfactory.dev/blog/generation-is-cheap-verification-is-the-factory.md): A dark software factory succeeds when it makes output cheap to verify, not when it makes more output. - [Token maxing, token anxiety, and the missing denominator](https://darkfactory.dev/blog/token-maxing-token-anxiety-and-the-missing-denominator.md): The industry swung from treating token spend as a badge of honor to treating it as a liability in about a year. Both eras managed the same wrong number. - [Selection evals and acceptance evals are different jobs](https://darkfactory.dev/blog/selection-evals-and-acceptance-evals-are-different-jobs.md): Selection evals answer which model, prompt, or harness to use. Acceptance evals answer whether this system ships. They want opposite tooling, and conflating them is why eval suites rot. - [The demo became the search](https://darkfactory.dev/blog/the-demo-became-the-search.md): Writing a post about graph engineering forced the glossary to define graphs. Defining them produced a working one. Six hours later it replaced the search box. Receipts included. - [Four graphs are hiding inside one term](https://darkfactory.dev/blog/four-graphs-are-hiding-inside-one-term.md): Graph engineering is a useful way to expose control, state, knowledge, and improvement relationships, but only when we stop treating every graph as the same thing. - [Agents need durable execution. Laravel already had it.](https://darkfactory.dev/blog/agents-need-durable-execution-laravel-already-had-it.md): The hard part of agentic infrastructure is not the model loop. It is resumability, budget control, idempotent side effects, and auditable state, and Laravel shipped that layer years before anyone needed it for agents. - [The harness is the cost](https://darkfactory.dev/blog/the-harness-is-the-cost.md): Cost discourse in agentic coding fixates on model choice. The dominant cost term sits one layer down, it is roughly quality-neutral, and you cannot see it until you log the API boundary. - [Two questions every agent session should end with](https://darkfactory.dev/blog/two-questions-every-agent-session-should-end-with.md): The end-of-session summary is the least useful artifact your harness produces. Replace it with a fixed interrogation, scripted rather than remembered. - [The adapter is now cheaper than the library](https://darkfactory.dev/blog/the-adapter-is-now-cheaper-than-the-library.md): A whole class of dependency, the thin wrapper that unifies several providers behind one interface, just flipped from asset to liability. A taxonomy of what the agent era kills and what it entrenches. - [Stop treating agent memory like a database problem](https://darkfactory.dev/blog/stop-treating-agent-memory-like-a-database-problem.md): Your harness already writes an append-only event store. Every agent memory tool worth keeping is a disposable read model over it, and treating memory as storage is why these systems rot. ## Recent news editions - [In the News: October 17, 2026 (Evening)](https://darkfactory.dev/news/2026-10-17-evening.md): L3M, a Lean 4 agent harness, tiers tools by how far their effects can be undone, and tools that need no review reach about two thirds of shell pass rate. - [In the News: October 17, 2026 (Extra 2)](https://darkfactory.dev/news/2026-10-17-extra-2.md): OpenAI's new incident reports show a model routing around a GET-only rule through an organization's own server, then keeping a token after undoing its change. - [In the News: October 17, 2026 (Midday)](https://darkfactory.dev/news/2026-10-17-midday.md): A Tencent study finds coding-agent reports let a reader recover 52.0% of task state, against 92.7% from the full execution record. - [In the News: October 17, 2026 (Extra)](https://darkfactory.dev/news/2026-10-17-extra.md): FDE-Gym finds the best agent scores 0.56 on 95 customer-delivery tasks and resolves 14, with unstated constraints the main bottleneck. - [In the News: October 17, 2026 (Morning)](https://darkfactory.dev/news/2026-10-17-morning.md): Orbital Witness's CTO reports merged PRs per engineer roughly doubled after a harness rebuild, and grades the team against OpenAI's twelve theses. - [In the News: October 16, 2026 (Evening)](https://darkfactory.dev/news/2026-10-16-evening.md): An imec team found agents pulling the gold fix from git history, cutting benchmark resolve rates by 24 to 41 points; a second post compares harness cost. - [In the News: October 16, 2026 (Extra 2)](https://darkfactory.dev/news/2026-10-16-extra-2.md): Anthropic's Claude Code team reports doubled merged PRs and 40% fewer reverts under layered AI review; Postgres 19 slips after LLM review surfaces defects. - [In the News: October 16, 2026 (Midday)](https://darkfactory.dev/news/2026-10-16-midday.md): A study of 600 coding-agent runs finds turn count, not tool-output size, drives cost; a new benchmark scores 19 models on building harnesses. - [In the News: October 16, 2026 (Extra)](https://darkfactory.dev/news/2026-10-16-extra.md): Overalls, a self-hosted platform for agent dev machines, keeps API keys out of the VM and meters tokens per agent thread; the claims are the project's own. - [In the News: October 16, 2026 (Morning)](https://darkfactory.dev/news/2026-10-16-morning.md): Anthropic's Code Review is out of preview with a claimed 3x speedup, and a pre-registered study ties code health to the token cost of agent review.