Cory Ondrejka, CTO of defense-tech firm Onebrief and co-creator of Second Life, has published "The o16g Manifesto," a 16-principle framework he calls "outcome engineering." The term is styled as a numeronym — o16g, counting the 16 letters between the 'o' and 'g' in "outcome engineering," in the tradition of i18n and a11y. The central argument is direct: agentic coding shifts the production bottleneck from human bandwidth to compute cost, rendering the traditional backlog obsolete. In Ondrejka's framing, ideas should be rejected only for lack of budget, never for lack of developer capacity.
The manifesto's 16 principles split across two parts — intent definition and agentic infrastructure. The more provocative claims sit in Part I. "The Liberation" declares the backlog dead outright. "The Truth" reframes the success metric entirely: lines of code mean nothing; only <a href="/news/2026-03-15-validation-is-the-missing-layer-in-llm-agent-workflows">verified customer outcomes</a> count. "The Artifacts" argues that failed agent runs shouldn't be rolled back and forgotten — they're data to dissect. Part II addresses the machinery: <a href="/news/2026-03-14-agentlog-lightweight-kafka-like-event-bus-for-ai-agent-orchestration-via-jsonl">swarm coordination</a>, encoding mission as machine-parseable constraints, risk as a hard production gate, and continuous auditing of agent reliability.
Ondrejka's background at the Naval Academy and his current role building AI-augmented military planning software at Onebrief give the manifesto an operationally conservative character that distinguishes it from typical startup-adjacent AI thought leadership. Principles like "The Gate" — which demands that unmitigated risk stops the production line entirely rather than appearing in a report — and "The Validation," which opens with "Trust is a vulnerability," reflect zero-trust and go/no-go gate thinking borrowed directly from defense and intelligence contexts. The emphasis on documented reasoning chains, automated post-mortems, and codified intent reads less like aspirational best practice and more like institutionally mandatory doctrine for high-stakes environments.
The o16g.com site is itself a technical demonstration of the manifesto's multi-model philosophy: built with Astro, running Claude Opus 4.6 and OpenAI's gpt-5-nano and gpt-5-mini as LLM backends, with pipelines through Cloudflare Workflows. Each of the 16 principles links to a curated corpus of supporting stories ranging from 41 to 460 items. That's a substantial knowledge base — and for a framework positioning itself as doctrine rather than opinion, the sourcing matters. If the compute-cost bottleneck thesis holds, the manifesto's insistence on risk gates and automated auditing won't stay a niche concern for long.