Our mission
To help builders and businesses scale agentic workflows to deliver awesome customer experiences.
Solving this is not just an engineering problem and Hosho was founded on that conviction. We built the team around AI science, engineering and business because the problem needs all three.
The problem
We have lived this problem deeply ourselves: the last year in stealth embedded in customers’ engineering and product teams, working with global enterprises to transform with agentic customer service and sales, and coming from a decade of AI evals for space applications.
Why is this hard?
- Typical state of play: vibe checks, a few LLM judges nobody trusts, a first-gen eval tool tried and quietly abandoned, and approaches often based on legacy Software QA
- Defining good is harder than it sounds. For an agent, good is a judgment, not a passing test: on-brand, accurate, resolved, at the right cost. It differs for every business, and most teams have never written it down.
- Measuring it is a discipline most teams don’t have. The right eval for each workflow, calibrated so the team trusts it, and kept honest while the product ships faster than ever.
- Even with measurement, the fix stays unclear. Causes span infra, orchestration, context, prompts and models. Finding the root, changing one thing, and proving it helped is where teams stall.
How Hosho helps
Hosho is the independent evaluation and fix layer for enterprise AI. The Hosho platform is used by leading US AI-native companies at the frontier of agentic workflows.
- Measurement you can trust. Calibrated deterministic and LLM judges that measure customer experience, output quality, and every agent, handoff and tool call in between. Anchored on what good means for your business, built on a decade of AI research.
- The full loop, closed.A world model of each agent, what was asked, what actually happened and what’s true, pinpoints the root cause of a failure, builds the fix as a pull request, and validates it held on live traffic.
- Major reduction in time and effort. Eval to triage to fix, automated and kept current as you ship, with no specialist hire to run it. Findings land where your teams already work: Slack, the IDE via MCP, Git, your database.
Why it’s us
The Hosho team pulls together the distinct capabilities required to solve this.
- Prateik — 10 years at Bain, including founding Bain Vietnam. Computer science engineer.
- Greg— 10 years at Bain & Co measuring and improving customer experience, 2.5 years at Salesforce working with enterprises implementing agentic workflows
- Nitish — 12 years in generative AI, including AI research and evals for space applications. Exited founder of a B2C AI product company
- Plus a team of highly experienced software and platform engineers building the Hosho platform.
Hosho’s early backers include investors, founders and executives who see the AI trust problem from inside global enterprises and AI labs every day.
Reach out
If you’re wrestling with how good your AI is really doing for your customers — whether you built it yourself or you use a vendor — we’d love to talk. Contact us.