25 Questions to Ask an AI Developer Before Starting a Project
A practical checklist for interviewing an AI developer before signing a contract
Readiness scoring, pass criteria, data checks, integration risk and ownership after launch
How this guide supports the broad Armenia AI specialist landing page with long-tail procurement intent
questions to ask an AI developer, AI vendor selection Armenia and production AI checklist
$ interview ai_developer --before-contract
> questions: 25
> score: outcome / data / integration / risk / ownership
> output: audit / prototype / production-readyHiring an AI developer before the project is clear creates a familiar problem: the conversation jumps to models, tools and demos before the company has tested whether the workflow, data, integrations and risk boundaries are ready.
The broad commercial intent belongs to the AI specialist in Armenia landing page. This article is narrower: it gives a practical checklist for teams that want to interview an AI developer, AI studio or AI contractor before signing a contract.
When to use this checklist
Use the checklist before a paid discovery, prototype, RAG system, AI automation, AI agent, LLM workflow or internal AI tool. It is useful when the team knows the business problem but has not yet converted it into a delivery brief.
The questions are not designed to produce a perfect score. They are designed to expose unknowns early:
- unclear business outcome;
- missing data access;
- hidden integration work;
- vague evaluation method;
- no human review boundary;
- weak ownership after launch.
If a vendor cannot discuss these topics in plain terms, the project is not ready for a production commitment.
Criteria and scoring scale
Score each answer from 0 to 3.
| Score | Meaning | Practical interpretation |
|---|---|---|
| 0 | Unknown | The answer is missing or speculative |
| 1 | Partial | The topic is understood, but evidence is weak |
| 2 | Usable | The answer is specific enough for discovery or prototype |
| 3 | Production-ready | The answer includes owner, data, checks and failure handling |
The goal is not to force every question to 3. A first audit can start with many 1 and 2 scores. A production build should not start while critical questions about data, permissions, errors and ownership remain at 0.
The 25 questions
| Area | Question | Pass signal |
|---|---|---|
| Business outcome | 1. What business decision or task should the AI system improve? | The answer names a workflow, not only a feature |
| Business outcome | 2. Who will use the result every week? | A real role or team is named |
| Business outcome | 3. What would count as a failed implementation? | The team can name unacceptable errors |
| Business outcome | 4. What is the smallest useful release? | Scope is smaller than the final vision |
| Data | 5. Which documents, CRM records, messages, files or APIs are needed? | Sources and owners are visible |
| Data | 6. Who is allowed to access this data? | Permissions are explicit |
| Data | 7. How often does the data change? | Update path is not ignored |
| Data | 8. What data must never be sent to a model or third-party API? | Sensitive fields are named |
| AI behavior | 9. Should the model answer, classify, draft, search, route or trigger actions? | The behavior is specific |
| AI behavior | 10. Does the workflow need RAG, fine-tuning, rules, n8n or ordinary code? | The vendor can compare options |
| AI behavior | 11. What examples will be used to test output quality? | Evaluation examples exist |
| AI behavior | 12. What should happen when confidence is low? | Fallback is defined |
| Integration | 13. Which systems must read data from the AI layer? | Inputs are mapped |
| Integration | 14. Which systems can the AI layer write to? | Write permissions are controlled |
| Integration | 15. Are webhooks, CRM fields, spreadsheets, queues or APIs already documented? | Integration contracts exist or are planned |
| Integration | 16. What latency, rate limits or uptime constraints matter? | Operational constraints are discussed |
| Human review | 17. Which decisions require human approval? | Approval gates are named |
| Human review | 18. Who reviews drafts, exceptions and edge cases? | A role owns review |
| Human review | 19. What should be logged for audit and debugging? | Logs are part of the design |
| Risk | 20. What are the main privacy, security and compliance risks? | Risk register exists |
| Risk | 21. How will prompt injection, bad source data or hallucination be handled? | Failure modes are concrete |
| Risk | 22. What rollback or manual process exists if AI fails? | Manual fallback is preserved |
| Delivery | 23. What artifacts will be delivered besides the demo? | Repo, docs, tests and deployment notes are named |
| Delivery | 24. Who owns maintenance after launch? | Ownership is not vague |
| Delivery | 25. What should be audited before expanding scope? | Next review gate is explicit |
How to fill it in
Run the checklist with the internal team first. Then use the same file during vendor calls. A strong AI developer should improve the checklist, not skip it.
For each question, write:
- current answer;
- evidence or missing input;
- score from 0 to 3;
- owner;
- next action.
This turns a sales conversation into a delivery conversation. It also prevents the team from mistaking a polished prototype for a production-ready system.
How to interpret the result
Use the total score only as a rough signal. The distribution matters more.
| Result | Interpretation | Next step |
|---|---|---|
| Many 0 scores in business outcome | The team is not ready to buy development | Run a short AI audit |
| Many 0 scores in data and permissions | The project may fail before AI quality is tested | Map sources, access and privacy rules |
| Weak integration answers | The demo may not connect to real operations | Create an integration brief |
| Weak evaluation answers | Output quality will be argued by opinion | Build an evaluation set |
| Weak ownership answers | The system may become unmaintained after launch | Define support, logs and handoff |
For a production AI project, pay special attention to questions 6, 8, 12, 14, 17, 19, 20, 21, 22 and 24. These control the risk around data, writes, failure handling and ownership.
Downloadable checklist template
Use this Markdown checklist as the original proof artifact.
| # | Question | Current answer | Score 0-3 | Evidence | Owner | Next action |
|---|---|---|---|---|---|---|
| 1 | What business decision or task should the AI system improve? | |||||
| 2 | Who will use the result every week? | |||||
| 3 | What would count as a failed implementation? | |||||
| 4 | What is the smallest useful release? | |||||
| 5 | Which documents, CRM records, messages, files or APIs are needed? |
The full Markdown file is available as a project asset: AI developer questions checklist.
What to do after the check
If the checklist exposes unclear scope, start with an audit. If the business workflow is clear but the data is messy, start with data and integration mapping. If the workflow, data, permissions and review gates are clear, the team can scope a prototype or first production-safe release.
For broader service context, use the AI specialist in Armenia page. For proof, review the case studies. For neighboring procurement criteria, compare the guides on choosing an AI developer, freelancer vs AI studio and AI specialist criteria.
Checked and updated
Checked on 2026-07-08 against the aicoding.am content plan, existing service pages, public case-study layer and current local article cluster. The article avoids self-awarded rankings and keeps broad local AI intent with the dedicated landing page.
Where This Applies
AI developer interview, procurement review and first production-safe scope
This article is useful when a company in Armenia wants to evaluate an AI developer before committing to an AI automation, RAG, agent, LLM workflow or internal tool project.
- Founders preparing the first vendor conversation.
- Operations teams turning an AI idea into a scorable project brief.
- Companies that need a checklist instead of broad claims about AI capability.
project_readiness = outcome_clarity
+ data_access
+ integration_contracts
+ evaluation_cases
+ ownership_after_launch;
if (score.data == 0 || score.permissions == 0) require("discovery_audit");
if (writes_to_live_systems) require("human_approval_gate");