Press & stories

Quick-lift facts, story angles, and assets for journalists, newsletter writers, and AI researchers writing about Cooli.

Boilerplate

Cooli AI is a small consultancy that designs, builds, and operates production AI systems for client teams — strategy through implementation, with the evals to back it. Cooli works across four lanes: agentic systems, LLM applications, integration & evals, and AI strategy. Engagements typically start with a one-week scoping and end with a system in production with monitoring.

Cooli also runs a public Lab — Sprout and Mulch — two GitHub-hosted experiments where AI agents build real software in the open. Sprout takes human-authored issues and lets a Claude agent build, judge, and commit the result. Mulch is a bot-only contribution zone where an autonomous gatekeeper auto-merges or roasts incoming PRs.

Story angles

"Delegated authorship in public"

Sprout and Mulch are working examples of AI as a code contributor with rules — refusal heuristics, sacred-file protections, and rate limits. Useful frame: what does it look like when AI agents are first-class contributors to a real GitHub org with real CI?

"How small consultancies ship AI"

Most AI projects fail in the gap between prototype and prod. Cooli's working pattern — eval set on day one, working code by week one, kill criteria up front — is a counterpoint to the prevalent "build a demo and hope" path.

"Agents in the wild"

Mulch's gatekeeper is an autonomous Claude validator that roasts non-conforming PRs. The roasts themselves — short, dry, written in a defined voice — are a small case study in AI tone control at production scale.

"AI-native marketing site"

Cooli.ai's own intake is a Gemini-driven chat. The service detail pages have inline industry-to-idea sketching. The site is a product of its own services — useful for any "show your work" angle.

Quotable lines

"We don't ship slide decks. The deliverable is working code with the eval set that proved it works." — Cooli, on engagement style
"Most agent projects bottleneck on tool reliability and observability — not on the model. We built around the boring parts first." — Cooli, on agentic systems
"The Lab is a way of being honest in public. The rules are on display, the failures are on display, and the wins are at cooli.ai/sprouts/." — Cooli, on the public Lab

Lab links

🌱 Sprout

Humans speak, AI builds. Open an issue → autonomous Claude builds → live page.

cooli.ai/sprouts/

🤖 Mulch

Bot-only contribution zone. Autonomous gatekeeper roasts or auto-merges.

cooli.ai/mulch/

Source — Cooli Lab

All Lab repositories, workflows, and validation scripts.

github.com/Cooli-Lab

Logo

PNG with transparent background. Use freely with credit.

/logo.png

Fact sheet

Founded: 2026. Headquarters: remote. Focus: production AI consulting + engineering. Stack-agnostic: Anthropic, OpenAI, Google, open weights — model selection is downstream of evals. Engagement length: 1-week scoping to multi-month builds. Public Lab: two experiments, all source open, all rules public.

Press contact

Email hello@cooli.ai with "Press" in the subject. We'll get back within 1 business day.

For ongoing coverage of Cooli Lab activity, the gallery pages are live: sprouts · mulch.