Alone among its peers, Kiewit ships a live llms.txt (the file AI assistants read first) and an Organization schema (the invisible labels search engines and AIs read): the front door for AI readers is open. What's behind it is still a passive content index: no verified work, no proof. This concept adds the room: proof-of-safe-completion for your own self-perform craft: hazard-gated packets, photo+GPS+signature proof, settled to a permanent on-chain receipt. Nobody Gets Hurt, now verifiable.
Kiewit already publishes a live llms.txt, ahead of every peer. But it's still a list a machine can read, not something a machine can verify.
We drafted the missing piece for free: a way to prove a specific craft task was done safely, closed with a permanent public receipt. The only ask is one real receipt, funded by us, for about $500, no meeting required.
llms.txt (HTTP 200, 2,495 bytes, Yoast-generated), ahead of every peer. But it's a passive SEO content index: no agent actions, no capabilities, no verification layer; agent-card.json is a 404./careers/ page, the front door to a 34,500-person self-perform craft workforce, carries only the Yoast graph. Zero JobPosting schema; agents can't read open craft roles as jobs.Baseline credited first: Kiewit is the most agent-forward of its peers. Then four evidenced findings: the empty room behind the open door, the schema-less craft careers page, the prose-only safety record, and one hygiene note, each with the exact URL, HTTP status, and re-check command.
Read the briefTwo acts in Kiewit black & yellow: Act 1, an agent reads the live llms.txt and hits the wall (content, no proof). Act 2, a self-perform craft task becomes a work packet that cannot be authored without passing a machine hazard gate, closed with photo+GPS+signature proof, settled to an on-chain receipt owned by the employee-owners who did the work.
Concept files that upgrade the passive llms.txt into an actionable surface: an agent card and MCP tool schemas (the files that let an AI assistant use you correctly) for verified craft safety history and structured craft jobs, and a concierge persona. Drafts, ready to hand to Kiewit's own team.
construction_site_safety_walk, stream_bioassessment, and watershed_outfall_screen lanes (the last two map directly to Kiewit's own Salton Sea and "Cleaning the Creek" projects); proof-checked packets (photo+GPS+signature); on-chain settlement on Base: EWP contract 0x76c17C…A14B, first settled packet workId 14; agent-payable tools over x402 (a web-native way for AI agents to pay per use) (work-packet $0.10 · work-proof-check $0.01); and a live end-to-end agent walkthrough: the building-agent golden path.
work-packet program answers HTTP 402 with a standard x402 payment schema; the agent signs a USDC authorization and pays $0.10. A runnable client with a dry-run default is published at vealth.net/recipes.html.Why this isn't scary: the agent cannot do anything invisible here. Every action embeds its own receipt: permanent, public, verifiable. That is more accountability than most enterprise software offers, not less, exactly what a Nobody Gets Hurt culture wants.
The bet: AI competence is scaling, and Kiewit has already bet on it, standing up Kiewit Technology Group and in-house AI tools like KADE and ADAPT, with Chief Engineers integrating AI into real workflows. You already opened the llms.txt door; this is the safety-verification rung behind it. Being early costs $0.10 and one afternoon. Being late means the proof-verified track records of your own craft belong to someone else's platform.
work-packet x402 call against our live endpoint today (vealth.net/.well-known/x402). Inbound-only, no outreach. [email protected].
