A Practical Guide to Choosing a Managed Engineering and AI Partner

Picking a managed engineering partner is a decision that quietly shapes the next 12–24 months of your roadmap. Get it right and you effectively gain a senior team overnight. Get it wrong and you spend a year managing a vendor instead of building a product. The questions below have surfaced almost every misalignment we've seen — on both sides of the table. Use them literally on your next call.

1. Who exactly is on the team?

Ask for named engineers with real seniority, LinkedIn profiles, and years-of-experience numbers — not a generic 'pool of over 200 engineers'. A serious partner will tell you who is billable to you, what percentage of their week you get, who backs them up when they're on leave, and what their background actually is. If the answer is vague, the seniority is probably vague too.

2. What's the mismatch policy?

Any long engagement will have at least one bad fit. Good partners commit up front to replacing an engineer inside a defined window — we do it in seven calendar days — with no debate and no invoice pressure. If replacement is treated as 'a conversation to have', it's a red flag: you're negotiating from a weak position the moment you need the change.

3. Who owns delivery?

Insist on a single accountable owner across product, engineering, deployment, and operations. The moment ownership is split across four handoffs — 'that's a product question, that's a DevOps question' — everything slows down. A good managed partner gives you one person you can escalate to, and that person has the authority to move budget, scope, or people to fix the issue.

4. Where is AI actually running in your partner's own business?

A partner selling AI capability should be operating AI internally — for customer support, sales ops, code review, hiring screens. If they can't demo it live, they're reselling someone else's platform. Ask specifically: 'Show me one AI system your team built and now runs in production for yourselves.' The answer is instructive.

5. Is the pricing transparent?

Flat monthly rates per engineer or per pod, published up front, are healthy. Complex rate cards with change-order surcharges, tiered discounts, and 'strategic partnership' addenda usually exist for a reason — and it's not your reason. If you can't predict next quarter's invoice within 5% today, walk.

6. What does exit look like?

The best partners make it trivial to walk away — full source code in your repo, full documentation, full credentials, full ownership of accounts and cloud infrastructure. That confidence is the entire point. If exit clauses feel punitive, or if you're being asked to run infrastructure on their accounts, you've locked yourself in without knowing it.

7. Can you speak to two current customers?

Not case studies — customers. On a call. If a partner can't name two current clients willing to take a 20-minute reference call, either the relationships aren't as good as advertised or the tenure isn't as long as claimed. Both matter.

8. How do they handle security incidents and disclosure?

Ask for the written incident-response process, the disclosure SLA, and whether they've had an incident in the last 24 months. A partner who says 'never' either has a very short history or a very selective memory. What you want to hear is a specific example, what happened, and what changed afterwards.

9. What's the retention rate for senior engineers?

The core promise of a managed engagement is continuity. If a partner's average senior engineer stays 14 months, you'll rebuild your team every 14 months whether you want to or not. Get the number in writing.

10. Can you see a real architecture review from an existing client?

Redacted is fine. What you're looking for is depth of thought — do their architecture reviews read like the work of a senior team, or like a template with your name search-and-replaced in? This is often the most revealing artefact in the whole evaluation.

How to actually run the evaluation

1. 1.Shortlist three to five partners and put all of them through the same ten questions.
2. 2.Score answers relative to each other, not against an ideal you invented.
3. 3.Have your engineering lead in every call — commercial-only calls hide most of the signal.
4. 4.Do a paid 2–4 week pilot before signing a long contract. Real work exposes real dynamics.
5. 5.Talk to the reference customers before, not after, you sign.

How CogneticAI answers these

We publish flat pod pricing, commit to seven-day engineer replacement, give clients full ownership of code and infrastructure from day one, and can name every senior engineer on our bench. We're not the right fit for every buyer — but the questions above will tell you honestly whether we are for you.