AI inside the product you already built for them.
Your client does not want a separate AI tool. They want the app you built last year to draft the reply, summarize the thread, pull the fields off the invoice and file it in the right place. That is usually a sprint, not a rebuild.
Scope this with us“Could the system just write the first draft so my team is only editing?”
Two to four discrete features fit comfortably in one 14-day sprint, assuming we are working inside an existing, running codebase.
Features inside your codebase
Built in your repo, in your framework, matching the patterns your team already uses — not bolted on as a separate service.
Streaming UX that feels fast
Token streaming, optimistic states and graceful loading, because a 6-second blank spinner reads as broken.
Structured, validated output
Extraction and classification return typed, schema-checked data your app can rely on rather than free text you have to parse.
Cost and usage controls
Per-tenant limits and cost visibility, so your client's bill cannot surprise either of you in month two.
The parts that decide whether it survives contact with users.
- Schema-constrained output with validation and a repair path when the model returns something malformed
- Prompt versioning, so you can roll back a regression instead of trying to remember what changed
- Caching on repeated inputs — often the difference between a viable margin and a bad one
- Fallback models when the primary provider rate-limits or goes down
- Evaluation on a real sample before launch, not vibes on three test cases
- Latency budgets per feature, chosen against where the user is actually waiting
- Ship a chatbot the users never asked for because it was the easiest thing to build
- Parse free-text model output with regex and call it structured data
- Leave cost uncapped and let the client discover the bill
- Hardcode one provider so tightly that switching means a rewrite
Before you scope it.
Can you work in our existing codebase?
Yes — that is the common case. We work in your repo, follow your conventions and open PRs your team reviews. If your stack is unusual we will tell you on the call whether we are a fit.
Which model provider do you use?
We pick per feature based on cost, latency and quality, and keep the provider behind an interface so switching later is a config change rather than a rewrite.
How do we price this for the client?
We give you real per-run cost numbers from the build, so you can price with a margin instead of guessing and absorbing the difference.
Want to stop turning down AI projects?
Bring the scope you’re unsure about — an RFP, a client request, a half-quoted project. We’ll tell you what’s buildable, what it takes, and whether it fits in one sprint. No charge for the call.