https://gethuman.sh/articles/ Recent content in Articles on human - The Dark Software Factory for Claude Code Hugo en Mon, 25 May 2026 00:00:00 +0000 https://gethuman.sh/getting-to-the-dark-software-factory/ Mon, 25 May 2026 00:00:00 +0000 https://gethuman.sh/getting-to-the-dark-software-factory/ <aside class="tldr"> <span class="tldr-label">TL;DR</span> <p>A dark software factory ships work from intent to production with no developer touching it on the way. Darkness is per-unit — about which stages run unattended, not how much AI you use overall — and it needs three things from the infrastructure, not the agent: machine-derivable intent, machine-observable outcomes, and a machine-bounded blast radius. A qualifier routes each unit into a dark lane or leaves it to humans; lanes are defined by intent source and capability (maintenance, bug autofix, telemetry-driven, transformation, gardening), while concerns like passing tests and stable APIs are the floor every lane must clear. Run continuously, the qualifier becomes a forager that cherry-picks work, and that scoring function — not the agents — is the real product. As shipping costs collapse, the bottleneck moves from what you can build to positioning and attention, so the factory's real value is throughput plus restraint.</p> https://gethuman.sh/run-a-dark-factory-on-your-codebase/ Sat, 23 May 2026 00:00:00 +0000 https://gethuman.sh/run-a-dark-factory-on-your-codebase/ <aside class="tldr"> <span class="tldr-label">TL;DR</span> <p>Yes, you can run a dark software factory on your own codebase today, and the recipe is lighter than the writeups make it look. Six categories already ship good results against the test infrastructure most teams already have — integrations and SDKs, migrations and refactors, CRUD on well-typed schemas, codegen from formal specs, infrastructure-as-code, and bug fixes with a reproducing test — together roughly 40–60% of what a typical SaaS team ships. The near edge (feature-flagged work, benchmark-driven performance) arrives within a year; the rings beyond it need Level 6 self-optimisation and more mature telemetry. Start by running one ticket end to end and shipping it, because the harness tuning and intent-writing compound with every ticket. human is the harness for that today-list.</p> https://gethuman.sh/beyond-dark-factory/ Sat, 11 Apr 2026 00:00:00 +0000 https://gethuman.sh/beyond-dark-factory/ <aside class="tldr"> <span class="tldr-label">TL;DR</span> <p>The first piece stopped at Level 5, the dark software factory. Two more levels are already visible from here. Level 6 is the self-optimizing factory: the harness measures itself, tunes itself, and compounds. Level 7 is the autonomous ideation factory: the factory reads customer signals and generates its own tickets. Each level pulls a loop that used to be human inside the machine. The meta-work is the new frontier, not code-writing. The scarce resource at Level 7 is taste, and taste does not move inside the factory. human is a Level 5 harness today; the next step for it is Level 6.</p> https://gethuman.sh/dark-software-factory/ Wed, 08 Apr 2026 00:00:00 +0000 https://gethuman.sh/dark-software-factory/ <aside class="tldr"> <span class="tldr-label">TL;DR</span> <p>A dark software factory is an autonomous software delivery pipeline where AI agents write, test, and ship code from specifications, and no human reviews the diffs. The name comes from lights-out manufacturing, where robots make things in factories with the lights off because no one is on the floor. The term was applied to software in early 2026 by Dan Shapiro. BCG Platinion and i-SCOOP picked it up as the category label for Level 5 AI-assisted coding, and StrongDM is the most-cited proof-of-concept. Humans are not removed from this model. They move upstream and downstream: intent, policy, outcome review. The middle, the actual writing and reviewing of code, is machine-driven.</p>