What “We Use AI” Actually Means (and When It’s a Lie)

"We use AI" became a client expectation, not a differentiator. So agencies started including it in every pitch without restructuring anything.

The minimum version is an engineer who has GitHub Copilot installed and uses it for autocomplete. The claim is technically true. The productivity impact is minimal.

The maximum version is a team that uses Claude or GPT-4 for architecture review, Cursor for code generation, AI test generation, and automated documentation — with senior review at every stage.

The same three-word sentence covers both. The only way to find out which one you are dealing with is to ask specific questions and require specific evidence.

Code generation: AI writes the first draft of a function, component, or module. A senior engineer reviews, tests, and refines. This is the highest-impact use case — it eliminates the rote scaffolding work that consumed junior developer time.

Schema and architecture design: using Claude or GPT-4 to review database schemas, API designs, and system architecture before implementation catches structural problems cheaply.

Test generation: AI writes unit and integration tests for existing or new code. What previously took 2-3 days of a QA engineer's time takes 2-3 hours.

Documentation: inline comments, README files, and API specifications are generated as code is written — not added as a post-project afterthought that never actually happens.

Code review and bug diagnosis: AI-assisted review catches common errors, security issues, and performance problems before code is committed. Not a replacement for senior review but a fast first pass.

Ask which specific tools they use — not "do you use AI" but Claude or Cursor or Copilot, and at which exact stage of the build.

Ask who reviews AI-generated code before it is committed. The right answer is a named senior engineer with a defined review process, not "the team reviews it."

Ask for a recent project timeline: how long from first commit to production deployment on a comparable scope? Real AI-native delivery shows 6-8 weeks for MVP scope, not 4 months.

Ask to see a code sample or a sanitised walkthrough of a recent module. An AI-native team can show you work. A team with AI only in the pitch cannot.

Ask what happens when AI generates incorrect code. The right answer is a specific QA and test process, not "we catch it in review."

Question one: Walk me through your last sprint. What did AI generate and what did engineers write manually?

Question two: What is your AI-to-judgment ratio — meaning what percentage of code is AI-generated vs human-authored?

Question three: Can I see a deployment from a recent project? Not a screenshot — a live URL.

Question four: What is your fixed-price confidence based on? AI-native teams quote fixed prices because they can predict timelines accurately. Teams without genuine AI integration cannot.

Question five: What does your test coverage look like at handover? AI-native teams deliver test suites automatically. Traditional teams often deliver minimal tests.

A written scope before any code is written — AI-native teams use AI to refine and validate scope, catching ambiguities early.

Delivery in 6-8 weeks for MVP scope that traditional teams quote 3-5 months to ship.

Test coverage at handover: unit tests, integration tests, and at minimum one end-to-end test per critical user flow — generated and refined during the build, not added at the end.

Documentation that is current as of handover: not a promise to document later.

A fixed price that the agency commits to — because their AI-assisted workflow gives them the timeline confidence to make the commitment.

Claude for architecture review, schema design, and complex logic: every significant design decision gets an AI-assisted second opinion before implementation.

Cursor and Claude Code for code generation: senior engineers direct what AI generates and review every output before commit. No raw AI code ships.

AI test generation: test suites are generated alongside code, not after delivery.

Documentation generated as code is written: README files, inline docs, and API specifications are current at handover.

Nastrum AI's AI-native workflow delivers a typical mobile or web app MVP in 6-8 weeks at fixed price — not because we cut corners, but because AI compresses the rote 70% that used to consume most of the timeline.

"We use AI" is a claim that needs evidence behind it. The evidence is specific: named tools, specific workflow stages, timeline data, and live products.

The agencies that have genuinely restructured around AI tools can answer every question in section four immediately and in detail.

The agencies that added AI to their pitch without restructuring their workflow will deflect, speak in generalities, or show you mockups rather than working products.

The stakes are high: genuine AI-native teams deliver in 6-8 weeks what traditional agencies take 3-5 months to ship on the same scope. That difference is your time to market, your budget, and your early user feedback loop.