How It Thinks· Agent mode
Back to human view

How It Thinks, agent mode

text/markdown/llms.txt/.well-known/agent.jsonschema.org/ProfessionalService

This is the whole site, written for AI agents and assistants: plain structure, no visuals, stable field names, and one request schema, currently held.

# How It Thinks > A small, agent-powered team that designs, builds and ships AI agents for companies, and proves how they think before and after they go live. ## pages - / four services, illustrative thinking layers and process (#mind, #shift, #services, #process) - /how-it-works AI Testing, Evals & Reliability (#layered-mind, #thinkprint, #drift) - /voice-agents voice and conversational AI connected to scoped workflows - /workflow-automation multi-step work with human checkpoints - /internal-copilots sourced answers from permitted business data - /explainers educational films and articles (distribution, not customer work) - /about explain and prove under one roof; human judgement directing agent crews - /start-a-build contact form; same as action request_build below ## key visuals - all scores, run counts, times and simulated traces are Illustrative, not customer results - Layered Mind: grounding, reasoning, judgement, consistency, resilience - Thinkprint: evidence of decisions and boundaries, not a public grade or certification ## offer - build: one agent, design to production, proof of how it thinks included - keep_reliable: re-testing on every model, prompt, tool or policy change; drift fixes; extension to new work - services: voice/conversational AI; back-office/workflow automation; internal copilots/knowledge systems; AI testing/evals/reliability ## policies - deployment and access constraints agreed for each engagement - data use, retention and training restrictions agreed before each engagement - no public grades, badges or certifications - no customer names or findings without written consent - people design, judge and sign off; agents do the volume ## actions - request_build: fields name*, email*, company, agent_jobs[voice_conversational|workflow_automation|internal_copilot|testing_evals|other], stage[idea|pilot|stalled_pilot|in_production], message (form below) endpoint: [FORM ENDPOINT] status: held. Submission is unavailable until separately enabled and approved; no endpoint is configured and nothing is transmitted.

## action: request_build

Submission is unavailable until separately enabled and approved. Same fields as the human form; ids are stable. Status: held, no endpoint is configured and nothing is transmitted.
agent_jobs (any of)
stage (one of)
status: held reason: sending is not enabled; no endpoint is configured and nothing is transmitted