Real product. Real cost moat. No developer brand.

Named public customers. Three products shipping weekly on owned compute — but no community, no content engine, an empty blog. That gap is the entire DevRel opportunity.

The lead is perishable.

The category is being defined right now — without Tenki in the conversation.

• GitHub cut hosted-runner prices ~39% — compressing the headline savings
• Blacksmith ($13.5M, GV) + Depot ($10M Series A) — better-capitalized pure-plays, louder brands
• CodeRabbit owns AI-review adoption mindshare
• E2B / Modal / Daytona racing for agent-sandbox mindshare — but all chasing the short-lived sandbox

Nick Hansen's framing: "we need to make sure we are the winner or one of the winners."

The thesis, in Tenki's words.

"Compute is the new digital oil. As global demand accelerates, scalable and tradable compute is set to become one of the most valuable markets of our lifetime."

Tenki was born inside Luxor — profitable, SOC-2, "compute as a commodity" — extending that thesis from mining into the agent era. The dev products are the wedge. Luxor's owned compute is the supply.

Three products that are secretly one.

Runners, Code Reviewer, and Sandbox all run on the same Firecracker microVM substrate.

• Runners — drop-in GitHub Actions, the cash wedge
• Code Reviewer — AI PR review, leading recall
• Sandbox — disposable microVMs for AI agents

In the agent era they converge: where an agent writes, runs, and gets its code reviewed in isolation. No competitor owns the whole loop.

Claim the category no one else is claiming.

Long-running, self-improving agents.

• E2B, Modal, Daytona optimize short-lived sandboxes — spin up, run, tear down
• The 2026 frontier is long-running, self-improving agents that persist across sessions
• Tenki's persistent volumes + snapshot/restore + Firecracker microVMs fit that shape exactly

The incumbents are racing for a category that's already crowded. This one is wide open — and Tenki is already built for it.

Four layers — and a model-provider land grab.

• Layer A · Primitives — Firecracker, gVisor, Kata. Tenki builds on Firecracker.
• Layer B · Platforms — E2B, Modal, Daytona and Tenki. Where harness builders shop.
• Layer C · Embedded — Cursor / Devin / Replit, sandbox as a feature.
• Layer D · Model-provider managed agents — Claude Managed Agents (Apr 2026) collapse model + sandbox + harness + state into one API.

Layer D can erase the Layer-B buying decision. So Tenki specializes where it's weakest: long-running agents, the integrated loop, owned-compute price — and MCP/harness-agnostic by design.

Own sandbox-first mindshare. Win the first five minutes.

• Most rivals treat the sandbox as an afterthought bolted onto a broader compute or dev-env product — Tenki can lead with it
• The native Sandbox ADE app is a real edge — first-run is trivial vs the hard self-host path of container-first rivals like Daytona
• Pair sandbox-first brand with owned-compute price and easy native UX

The Day-1 deliverable: a First-Five-Minutes audit — get time-to-first-successful-sandbox under five minutes.

Recall is the easy half. Trust is the hard half.

AI-review recall — ~2× the next best. But precision trails at 29.9%. The honest gap is the most interesting unsolved problem in the category — and our content engine. Luxor's own value: "honesty over kindness."

One north star.

Weekly active workspaces running ≥2 Tenki products.

A green CI run and an enabled Code Reviewer. Or runner + Sandbox. PMF (installed) + activation (a successful run) + stickiness + the multi-product moat (≥2). One number, all the answers.

Seven initiatives.

A proven 7-initiative framework — each program bolted to a SWOT lever.

• 1. Tenki Benchmark Series — the trust engine
• 2. Agent-Era Reference Architectures — Sandbox + BYO-agent
• 3. The 2-Minute Migration Engine — the runner motor
• 4. Community + Content Engine — the missing OS
• 5. Design Partner Cohort — manufacture the references
• 6. Tenki for OSS + Champions + Hackathons
• 7. Strategic Channel & Ecosystem Integrations

1–2 · Benchmarks & reference architectures.

• Benchmark Series — quarterly + per-model-release, always reproducible (open harness). Review recall and precision vs CodeRabbit, Greptile, Copilot, Cursor, Devin; runners $/run; sandbox boot latency. Publish where Tenki loses. Attacks W2, S5, T7
• Agent-Era References — Claude Code + Sandbox (Anthropic), Codex + Sandbox (OpenAI), Cursor / Devin / Cline. Repo + 5-min video + guide + cost calculator. "Give your agent root without giving it yours." Attacks S6, O2, O4, T4

3–4 · Migration & community.

• 2-Minute Migration Engine — Migration Wizard storytelling, honest comparison guides (vs GitHub, Depot, Blacksmith, Namespace), public cost calculator, first-five-minutes onboarding overhaul. Lean into the GitHub-pricing moment. Attacks S1, O1, W6, T2/T6
• Community + Content Engine — Discord (per-product + per-agent), weekly office hours, weekly build-in-public stream, biweekly newsletter, coordinated HN/Reddit/X launches, podcast guesting. Attacks W3, W8

5–7 · Cohort, OSS & channels.

• Design Partner Cohort — 12 weeks · 8–10 companies · 4 batches/yr. White-glove for case studies, quotes, co-talks, logos. Quarterly Demo Day livestream. Attacks W1, W8
• Tenki for OSS + Champions + Hackathons — free Runners + Reviewer for public repos; Champions program; "Agent Camp × Tenki" series; OSS PRs with prompt history. Attacks W1, W3, W5, O7
• Strategic Channel Integrations — GitHub Marketplace; Anthropic / OpenAI / Cursor co-marketing; agent registries; the Luxor compute tie-in. Attacks O6, T2/T3/T4

Three phases, three marketing moments.

• Phase 1 · Mo 1–3 — Foundations. Instrument the funnel · first-five-minutes overhaul · Discord + office hours + build-in-public · Benchmark Drop #1 · migration guides + calculator · Cohort #1 opens.
  Milestone: 500+ engineers, 1 benchmark live, time-to-first-run improving.
• Phase 2 · Mo 4–6 — Agent-era push. Sandbox references · Anthropic/OpenAI co-marketing · Agent Camp #1 · Cohort #1 Demo Day · Benchmark #2 · Tenki-for-OSS.
  Milestone: 1,500 weekly active workspaces, 8 case studies in flight.
• Phase 3 · Mo 7–12 — Scale + enterprise. Flagship events · Luxor compute narrative · Champions · Cohorts #2–3 · Benchmarks #3–4.
  Milestone: 3,000 workspaces, 20 references, 150 paid accounts.

Content + events highlights.

• Build-in-public — weekly stream fixing a live OSS bug with Tenki (the JD requirement, in motion)
• Benchmark series — the AI-reviewer teardown, then runners + sandbox head-to-heads, on a publishing cadence
• "How I ship with Tenki" — weekly community posts; biweekly newsletter; podcast guesting (Latent Space, AI Engineer, ChangeLog)
• Agent Camp × Tenki — own hackathon series on the dabl.club event engine; devs build on Sandbox
• Flagship presence — AI Engineer World's Fair, GitHub Universe, KubeCon (~$115K/yr event budget)

An audience on day one.

developers at dabl.club (36–40% open rate), pivoted hard into AI agents in early 2025. An attributed acquisition channel feeding Tenki-owned community, benchmarks, and events — instant top-of-funnel most startups spend a year building.

Three engagement options.

Substance is identical in all three. Only exclusivity and ramp differ.

• A · Full-time, all-in. Wind down Nebius over ~60 days, join as Head of DevRel. SF comp + equity.
• B · 90-day paid pilot → convert. A defined-scope pilot with OKRs and a conversion gate. The two-way-door path. Recommended.
• C · Fractional with OKRs. Monthly retainer + programs. The operating system, lighter commitment.

The pilot is Phase 1 of this plan, with hard metrics. You let the funnel decide.

Why Colin.

A builder-DevRel with the playbook and the audience.

• Cisco Meraki — built the developer program from scratch
• MetaMask — Head of DevRel, multiple L2 launches
• dabl.club founder — 86k developers, Agent Camp, 300+ AI events

Events I run: BuilderShip · ClawCamp · Self-Improving AI Agents Hackathon · Automate Engineering