Hunch designs, builds, and operates production agents for professional services firms. We embed for eight weeks, ship one agent live on our platform — fully traced, with your credentials encrypted per-client and the code in your repo.
We don't sell horizontal "AI for everyone." We embed in three industries we understand cold — and ship agents that match how your team already works.
Brokerages, owners, asset managers. We ship agents that handle the deal-flow grind so brokers can spend time in front of clients.
Search firms and high-growth talent teams. Agents that read a JD like a recruiter — and source, draft, and schedule like one.
PEOs and in-house People teams. Agents that own the policy-heavy, evidence-heavy work that drowns HR generalists.
Most "AI projects" stall because nobody owns the round trip from workflow audit to production operator. We do all three — and we don't bill for theatre.
Two weeks on-site with the operators who do the work today. We map every step, every exception, every system it touches — and leave with a directive spec (inputs, outputs, success criteria, edge cases) you'd sign off on. Prioritised roadmap, whether you keep working with us or not.
Six weeks to a production agent on the Hunch platform — Modal-hosted, instrumented with traces and evals from day one, your credentials encrypted per-client. Claude by default, or whichever model fits the job. The directive, the tools, and the deployment scripts all live in your repo.
Agents drift. We stay on as the operator: every run produces a full event trace, every failure becomes a PR that tightens the directive or fixes a tool, every redeploy is a one-command ship. Monthly reports on accuracy, throughput, cost. Optional — but most clients keep us through year one.
No multi-quarter "transformations." We co-locate, ship, and hand over. Most engagements start with one workflow and expand from there.
Fixed-fee. Senior team. One agent in production by week 8 or you don't pay the build fee.
Half-day workshop with the operators who do the work today. We leave with the top three candidates, ranked by ROI and feasibility, and a definition of done you'd sign off on.
We sit with the team, capture every step, every exception, every system the work touches. Surface what's actually slow, who actually owns it, and where the institutional knowledge lives.
One Hunch engineer + one of your operators per agent, paired daily. We instrument with traces and evals from day one. Weekly demos, weekly course-corrections, working software.
Rollout to a real cohort, side-by-side with the operators it's meant to serve. Then a runbook, an on-call rota, and a monthly review cadence — so the agent gets better instead of stale.
Promises about how we deliver, not retroactive case-study numbers. We'll replace these with engagement-level results as they land.
30-minute call. We'll tell you if we can help, and roughly what it would take. No deck, no sales engineer.
Book a strategy call