Methodology
These ranges are planning aids for a scoping conversation. They are not drawn from a counted set of invoices, and they are not a quote. A fixed price is confirmed after we see the repo and the workflow.
Key findings
The numbers we keep coming back to
The signals that shaped the rest of the analysis — each one is summarized with the methodology above.
Scope
What moves a quote
Auth, data access, payment webhooks, and whether the deploy will hold change the price more than the model name.
Written
Findings before a sprint
A useful audit names the blockers, ranks them, and says what a fix-up sprint would touch. It does not publish a rate on this site.
Quote
Fixed price, after the repo
We confirm a number once we have seen the code and the workflow. Illustrative ranges are for planning, not a bid.
Hidden
Work founders forget to budget
Evals, guardrails, observability, data cleanup, and a real on-call window show up after a demo looks done.
Data
Preparation is its own task
Retrieval and classification work depends on how clean the source material is. Treat data prep as scope, not a footnote.
Support
The quote includes the window
A production system needs a named response window. That window belongs in the statement of work, not as a published day count.
Inside the report
7 chapters of research
A scannable chapter guide of what the full write-up covers.
01
What changes by system type
Agents, copilots, retrieval systems, and classifiers are different scopes. A quote follows the workflow, the data, and the failure modes, not a published rate card.
- • Agents: tool use, permissions, and a way to stop a bad run
- • Copilot UX: where the model is allowed to act inside the product
- • Retrieval: sources, chunking, and what “wrong” looks like
- • Classifiers: labeled examples and a review path for mistakes
02
Regulated work is a different scope
Healthcare and fintech add access control, audit trails, and a written story for how patient or customer data moves. That is extra scope. It is not a certification we claim on this page.
- • Healthcare: identify sensitive fields before the build
- • Fintech: payments, webhooks, and who can refund
- • Ask for a security review when the feature touches patient data
- • Unregulated SaaS still needs auth, secrets, and a deploy that holds
03
Team shape changes the engagement
A senior review plus a small build team is a different offer from a large agency account team. Compare the written scope and the support window, not a regional multiplier we have not measured.
04
Who owns the handover
Price follows who reviews the code, who is on the hook after launch, and whether you receive the repo. Ask for that in writing before you compare bids.
05
Work that shows up after the demo
Evals, guardrails, observability, access control, data prep, on-call, and model spend are easy to leave out of a first estimate. Name them before you accept a fixed price.
06
Negotiation plays that actually move price
Which contract levers (milestones, source code ownership, maintenance windows) vendors will concede to and which they will walk from.
07
2026 forecast and pricing signals to watch
Agentic workflows are compressing classifier and RAG pricing while simultaneously expanding the top end of custom-agent pricing.
Full report
Request a copy of the 58-page report
The PDF is not published on this site. Tell us which report you need and we will follow up.