Graft AI

Graft will be helping f500's transform their legacy applications into operational knowledge basis for agents to use. Agents can interact with their legacy software and help them transition into agent native software, this is a shift we're seeing but companies fail because the data is in people's head as domain knowledge so we're building the infra to make it operational by agents.

Building a receipt-scanning feature right now and even at my tiny scale, the "verification" part is the hard bit — I ended up schema-validating every LLM response and failing closed on anything unparseable, because silently-wrong output is worse than an error. Curious how you handle the case where the workflow succeeds mechanically but the result is semantically wrong — is verification rule-based per workflow, or learned?

Operations as a map agents can walk is close to what I do per-client, structured facts an AI can't step outside of. Mine is one brand, yours is a whole company, and I suspect the hard part scales badly. How do you keep the map current when the operations change weekly?

One thing that would help us adopt this faster is a visual timeline view inside the operational map, so we can scrub through past workflow executions and see exactly where drift or failures happened. Right now troubleshooting at scale feels like guesswork.

the framing that the model was never the hard part is honestly accurate for most companies, the real question is whether the "living map" stays correct as workflows quietly drift over months, not just when the UI visibly changes.

spend months trying to get an agent to reliably click though our old internal ticketing system before giving up, this is the exact problem.

nothing in the pitch about workflows that need a human judgement call partway through, is that out of scope for now?

feels adjacent to Browser Use in spirit, just aimed at ERPs and desktop apps instead of the open web.

building stable tools instead of chasing every API makes sense, but that shifts a lot of ongoing maintenance onto graft's side to keep those maps accurate as companies change.

Hey PH 👋

I started Graft after realizing that the hardest part of deploying agents at work is not the model. It is the software around it.

Real companies run on ERPs, desktop apps, internal portals, spreadsheets, and years of operational knowledge that agents cannot reliably understand or use.

Graft began as a way to turn those interfaces into stable tools for agents. While building it, the idea grew into something bigger: a living map of how a company actually works. Graft learns the workflow, decisions, permissions, exceptions, and success conditions, then gives agents a safe way to do the work with approvals, audit trails, and verification built in.

When the underlying software changes, the agent-facing tool stays stable.

We are launching early because we want to learn from the people actually building and operating agents.

What is one system or workflow your agent still cannot reliably use today?