Most hiring guides read like they were written by agencies. This one is written by an agency — but we'll give you the questions that would eliminate 80% of the industry, including shops worse than ours. Use it honestly.
Before you talk to any agency: know your constraint
Write down one sentence: "I have [budget] and [timeline] to ship [one concrete outcome]."
Example: "I have $60K and 12 weeks to ship a patient-intake iPad app for my dental practice."
If you can't write that sentence, you're not ready. Go read our MVP framework, talk to customers, come back.
The seven questions that sort signal from noise
1. "Who specifically will work on my project?"
A good answer: a name, a photo, a LinkedIn link, years of experience, what they shipped before.
A bad answer: "our senior team" or "we'll assign someone based on availability."
If they can't name the person, they're treating you as fungible. You'll get whoever is available — which in most shops means whoever isn't billable.
2. "What's your code review process?"
Good answer: "Every PR reviewed by a senior before merge. CI enforces test coverage, static analysis, and security scans. Build fails on critical findings."
Bad answer: "We have governance" / "We follow best practices."
This question is especially important in 2026 because of AI-generated code quality issues. See AI-accelerated vs traditional.
3. "Can I see three case studies in my industry with named clients and outcome numbers?"
Good answer: they send PDFs or links within a day, with real numbers ($, %, uptime, users).
Bad answer: "We've done work in your industry but it's confidential." That's sometimes true, but if every case is confidential, they have no case studies.
4. "What's your fixed-price model, and when do you use hourly?"
Good answer: a clear range (e.g., "$25K–$150K fixed, hourly only for ongoing maintenance"), and they stick to it.
Bad answer: "Every project is different, let's get on a call to scope." Translation: they want time to feel out your budget and price against it.
5. "What happens if you miss the deadline?"
Good answer: they own it, you only pay for delivered work, and they offer a make-good.
Bad answer: "We've never missed" (they have) or "Scope creep usually causes delays" (deflecting).
6. "Can I talk to three clients from the last 12 months?"
A real agency says "yes" and makes intros within 3 days. If they hesitate, probe why. Common reasons: the client relationship ended badly, or there aren't many recent projects.
7. "What happens to the code on day one?"
Good answer: "Your GitHub org, your repo, you own it."
Bad answer: "We'll transfer it at project end" or "We host it for you." Translation: they're holding it hostage. Walk away.
The five red flags that predict a failed project
- No pricing information anywhere. "Contact us for pricing" is usually hiding either that they price-discriminate against wealthier clients, or they don't know what their work costs.
- Stock photos of smiling teams on the About page. Real agencies use real photos. Stock photography = no real people.
- Testimonials without verified platforms. A Clutch or G2 review is verifiable. A quote on their site is not. The first is worth 10× more than the second.
- Discovery calls before pricing. In 2026 you don't need a 45-minute call to get a price range. If an agency insists, they're either pricing per budget or not confident in their own estimates.
- A project manager as your primary contact. You want the engineer who's building it. PMs add a layer of telephone game between requirements and code.
What to ignore
- Awards pages. Design awards don't predict shipping quality. A Webby or ADDY means nothing for your operations dashboard.
- Number of employees. A 5-person senior shop outperforms a 500-person outsourcing firm on $50K projects.
- Years in business. 2017 and 2005 are the same from your perspective if both teams know what they're doing. New shops can be better than old ones.
- "Industry expertise." Unless it's highly regulated (healthcare, finance, defense), a good engineer picks up your domain in a week.
- Technology stack badges. "React expert" is not a differentiator; everyone who claims it uses React.
A two-week evaluation process that actually works
Talking to 3 shops, pick one, sign contract. Here's the structure:
Day 1 — Shortlist
Clutch + G2 + 2 referrals. Aim for 5 shops that have verified reviews in your budget range. Tier 1: shops ranked in our 2026 ranking.
Days 2–3 — Written briefs
Send the same one-paragraph brief to all 5. Don't take calls. See which ones reply within 24 hours with a written plan and a price range. Those are your finalists.
Days 4–7 — Three 30-minute calls
Use the seven questions above. Ask for references on each call. Ask for the engineer who'd lead, not the salesperson.
Days 8–10 — Reference calls
Talk to 2 recent clients per finalist. Ask: "Would you rehire them?" and "What was the one thing you'd change?"
Days 11–14 — Decide
Pick the one with: named senior engineer + fixed-price quote + verified reviews + code review process + clean references. Sign. Start.
The final signal
A good agency makes you feel smarter after the first call. A bad agency makes you feel like you need to explain your business in more detail. Pick the first.
Run your evaluation on us.
Send the same paragraph brief. We'll reply within 72 hours with an AI-generated estimate + a senior engineer who'd lead your build. No calls required until scope is priced.