The build-versus-buy question for agentic finance
It is a fair question, and one I respect: with capable models available by API, why not build your own finance agents instead of buying a platform? As the person who builds ours, let me give the honest version rather than the self-serving one.
What the demo hides
Wiring a model to your ERP and getting it to extract an invoice is a weekend project, and it works impressively in a demo. The distance from that demo to something you would trust with your ledger is enormous, and almost all of it is unglamorous: handling the messy 80% of cases, building the controls, the audit trail, the permissions, the human-approval flows, the error handling, and the monitoring that tells you when accuracy drifts.
The real cost of building
- Controls and audit infrastructure — the part finance actually requires — is most of the work and none of the demo.
- Integrations are a maintenance burden, not a one-time build; APIs change.
- Accuracy on your full distribution requires evaluation tooling you also have to build.
- Someone has to own it forever, including the night a model update changes behavior.
When building makes sense
If you have a genuinely unusual workflow, a strong platform team with spare capacity, and a tolerance for owning the controls yourself, building can be right — particularly for a narrow, stable, internal process. For the broad, regulated, audit-sensitive work across AP, AR, close, and procurement, the controls and integration surface is where buying earns its keep.
The model is the easy part. The trustworthy system around it is the product. Whichever way you decide, decide on that basis, not on the demo.