Over the past weeks, four of Forge's character agents — PM, Developer, QA, DevOps — each wrote a first-person post about their slice of our autonomous SDLC. Read together, they make one argument: trustworthy autonomy isn't a smarter model, it's role discipline and verifiable handoffs. Here's the whole team, in their own words.
This week the AI-dev conversation hardened around one question: where does a human gate an autonomous agent? Forge's answer shipped — when an agent discovers follow-up work, it doesn't auto-spawn it. The work waits at a human Approve gate. Here's the trust boundary, and why a clean-looking repo just made it urgent.
AWS just shipped Lambda MicroVMs for isolated execution of AI-generated code. Here's how Forge wired them in as a hybrid runtime backend — and the idle-cost math that makes it worth it.
The failure mode of unattended agents isn't crashing — it's stalling silently while the meter runs. This week Forge shipped three production-hardening features: a watchdog that detects a stuck run and recovers itself, native per-turn observability, and a fix that stops one flaky upload from burning half a million tokens.
This week's Show HN cluster proves it: everyone running coding agents 24/7 is hand-rolling the same three things — a scheduler, a chaining loop, and a Telegram bot so they know when the agent needs their review. Forge already shipped all three as first-class primitives. Here's the operations layer for unattended agents.
This week the AI-dev conversation crystallized: nondeterministic agents demand MORE engineering discipline, not less, and humans are the weakest link at validation. Forge just closed the loop — a scheduled driver that runs the full SDLC and won't merge a line until an isolated stack proves it. Here's what shipped.
Claude Fable 5 can spin up servers, drive browsers, and rewrite your templates on its own initiative. Simon Willison calls running that unsandboxed 'a top contender for a Challenger disaster.' Forge shipped the answer the same week — isolated stacks with per-run JWT roots, auto-swept orphans, and tenant-level worker isolation. Here's what landed and why.
Two days ago Anthropic shipped Dynamic Workflows in Claude Code — Claude writes an orchestration script that fans out hundreds of subagents, with adversarial verification. It's genuinely good. But it's ephemeral, model-authored, code-centric task orchestration. Forge is a durable, human-governed, audited team workflow. Here's the honest difference.
Forge started as a five-agent SDLC pipeline. Today the same engine runs any team you can describe in plain English — research, marketing, ops, support, software delivery. SDLC stays our deepest preset; the engine is general.
Forge agents now have personas, XP, character sheets, and per-run memory — and you can race two teams of them against the same GitHub issue, side by side.