If you’re running a startup with 5 to 50 engineers, everything you knew about building a tech team expired sometime in the last 18 months. The hiring funnel broke. The enablement model stopped working. The retention math shifted. And the old playbook — post job, screen CVs, run interviews, onboard, hope they stay — is failing at every stage.
I don’t say this from the sidelines. My company, A17, is a data engineering and AI services firm with around 50 people distributed across Russia, Armenia, Spain, Portugal, France, Thailand, and a few other places. We’ve hired across all these markets. We’ve made every mistake in this article at least once. Some of what follows is hard-won operational knowledge; some of it is validated by independent research; and some of it is still a bet we’re making that hasn’t fully played out yet. I’ll try to be honest about which is which.
This is not a step-by-step recipe. The whole point is that the environment is changing fast enough that any specific tactic has a shelf life measured in months. What I want to share instead are the underlying principles that seem durable — and some examples of how we and others have translated them into practice, which you can steal, modify, or ignore entirely.
The principles underneath#
Before we get into the three areas — finding, enabling, retaining — there are a few things I’ve come to believe that run through all of them.
AI is categorically different from previous tech waves. I’ve been in data and AI since 2005, back when we called it “database development” because nobody had better terminology yet. I’ve seen CRMs, ERPs, process automation, cloud migration, big data — all of them changed what companies built, but the way you managed people through those transitions was roughly the same. AI is different. It’s non-deterministic, it moves faster than any previous wave, and people relate to it emotionally in ways they never related to Salesforce or AWS. Your team has feelings about AI — excitement, anxiety, skepticism, existential dread — and those feelings directly affect whether adoption succeeds or fails. Managing through this transition requires a different kind of leadership.
People matter more than architecture. A great team will solve any technical problem you throw at them, even with a mediocre tech stack. A mediocre team will struggle even with a perfect architecture. This was true before AI and it’s more true now, because AI amplifies the gap between strong and weak engineers. The leverage of one excellent person with AI tools is enormous. The damage of one careless person with AI tools is also enormous. This makes every people decision — hiring, enabling, retaining — higher stakes than it’s ever been.
Your specific answers will be different from mine. We use questionnaires instead of CVs. We run paid test days. We have internal hackathons. These things work for us, in our context, with our team, in our markets. They might not work for you. What matters is whether you understand the problem well enough to design your own response. The tactics are examples. The principles are what transfer.
Finding: the funnel is broken, build a new one#
The core problem with hiring tech talent in 2026 is that AI broke the signal-to-noise ratio at every stage of the traditional funnel. About 70% of CVs are now AI-generated. Application volume per opening has roughly tripled. Take-home assignments get solved by Claude in minutes. Live coding interviews can be cheated with invisible overlay tools. The old sequence of CV screen → phone call → take-home → live coding → culture fit → offer was designed for a world where candidates were doing the work themselves and lying was hard. That world is gone.
The principle that seems to hold: move assessment toward real work as fast as possible. The closer your evaluation is to what the person will actually do on the job, the harder it is to fake and the more useful the signal. Everything else — CVs, cover letters, algorithmic puzzles, personality assessments — has been degraded to the point of near-uselessness by AI assistance.
How we’ve operationalized this: a short self-assessment questionnaire (binary and ternary questions, two minutes to complete) replaces the CV as an initial filter. One light screening call to confirm the person exists and communicates well. A brief technical conversation with a lead — not to stump them, just to see if they can talk shop. And then a paid test day where the candidate joins the team, gets a real task, interacts with real colleagues, uses whatever AI tools they want, and we watch how they actually work.
The paid test day is the anchor of the whole process. It costs roughly €300 for a senior engineer, compared to €5-6K for agency placement. Both sides see reality: the candidate sees your actual codebase and team dynamics, you see how they think, communicate, and handle ambiguity. No amount of interview prep or AI coaching can fake eight hours of working alongside people.
Is this the only way? Absolutely not. Linear does multi-day paid trials. Some companies run collaborative system design sessions. Others have gone back to in-person interviews entirely. Around 40% of startups now explicitly allow AI use during interviews, testing judgment rather than recall. The common thread isn’t the specific format — it’s the shift from evaluating artifacts (CVs, take-homes) to evaluating behavior in context. Figure out what version of that works for your team size, your budget, and your hiring volume.
I wrote about this in detail in Part 1: Your Hiring Funnel Is Broken. Here’s What Works Now.
Enabling: licenses are not a strategy#
You bought AI tools for the team. Great. Now what?
More than 80% of developers use or plan to use AI coding assistants. Cursor hit a billion dollars in ARR faster than any SaaS product in history. GitHub Copilot is in 90% of Fortune 100 companies. There is no shortage of powerful tools. The supply side is solved.
The demand side is where everyone gets stuck. A study by METR found that experienced developers using AI on their own repositories were 19% slower — while believing they were 20% faster. Real velocity gains across teams range from 10% to 70%, with a median around 40%. The difference between the low end and the high end isn’t which tool you use. It’s how your team adopts it.
The principle: AI adoption is a change management problem, not a procurement problem.
Three things seem to matter most.
First, team composition. Your people naturally split into roughly three groups: champions who explore everything on day one (~30%), a curious middle that’s pragmatic and builds sustained practice (~50%), and skeptics who push back and question everything (~20%). You need all three. A team of all champions is chaos — infinite experimentation, zero reflection, code quality in freefall. Skeptics catch errors that enthusiasts miss. Research from UCI found that people with negative AI bias demonstrate greater vigilance and achieve better outcomes. The diversity of attitudes toward AI is itself a form of quality control.
Second, leadership modeling. When managers visibly use AI in their own work, teams are 4× more likely to experiment. Four times. Your tech leads, your VPs of engineering, your CTO — they set the ceiling. If they’re not demonstrating competent, visible AI use and setting clear guardrails for acceptable use, adoption stalls at the surface.
Third, process redesign. Most teams bolt AI onto workflows designed for humans. Replace one step with an AI tool, move on. But the handoffs between human and AI create context gaps, broken documentation, and review processes that aren’t calibrated for AI output. Veracode found that 45% of AI-generated code introduces security vulnerabilities. Code churn is rising while refactoring approaches zero. The fix isn’t to ban AI — it’s to redesign your development process assuming AI is part of it from the start. How do requirements flow into the AI tool? How does AI output get documented? How does code review work when half the code was generated? These are the hard questions, and they’re specific to each team.
The deeper analysis is in Part 2: Your Tech Team Has AI Tools. Now What?
Retaining: the math changed, your engineers noticed#
Here’s the part most founders aren’t ready for. AI made your best engineers significantly more productive. Three people deliver what used to take five. And those three people know exactly what they’re worth now.
Ravio’s data shows AI-first startups at Series A/B operate 34% leaner but pay 36% more per individual contributor. Companies like Midjourney, Replit, and ElevenLabs have scaled to millions in revenue with fewer than 30 people. The “Super IC” is emerging: experienced professionals who combine strategic thinking with AI-augmented execution, eliminating entire support layers. This is the new competitive landscape for talent.
The principle: retention is now a function of growth opportunity, not just compensation.
Money matters — I’m not naive about that. We’re a bootstrapped company, no venture funding, and we’ve always had to compensate payroll constraints with other things. But even well-funded companies can’t just throw money at this. Betterworks found that 4 in 5 AI power users are actively job-hunting if they feel constrained in their AI usage. These are your most productive people. They leave not because of salary, but because they feel they’re falling behind.
What’s worked for us, and what I’ve seen work elsewhere: invest the time saved by AI back into growth. Not more sprint tickets — actual learning. Internal hackathons, show-and-tell sessions, sandbox time with new tools, shared prompt libraries. Our best hackathon wasn’t even planned by management; a couple of younger engineers just announced it and people showed up because it was genuinely interesting. That kind of organic experimentation does more for retention than any annual review conversation.
Hire one or two AI-savvy engineers and let them pull the rest of the team up. Internal upskilling compounds — every person who levels up becomes someone who can level up the next person. It’s vastly cheaper than trying to hire an entire team of experts at current market rates.
And have honest conversations about compensation before your competitors do. The engineers who’ve mastered AI-assisted development know their market value shifted. If you wait for the counteroffer conversation, you’ve already lost.
More on this in Part 3: Your Engineers Know They’re Underpriced. Now What?
What the playbook doesn’t cover (yet)#
A few things that emerged from research and conversations that I think will matter a lot over the next year or two, even though they’re not fully baked into anyone’s playbook yet.
The junior developer pipeline is hollowing out. Entry-level tech hiring at the 15 biggest tech firms fell 25% from 2023 to 2024. If AI handles boilerplate, what do juniors do? The answer probably involves supervising AI output and learning architectural thinking rather than writing basic code. But most companies haven’t figured this out, and the risk of having no talent pipeline in five years is real.
AI governance is coming whether you’re ready or not. EU AI Act provisions are in effect. U.S. states are fragmenting fast — Colorado, Texas, others. Even a 5-person startup needs basic policies: what AI uses are prohibited, what can’t be fed into public models, who reviews AI output for sensitive decisions. This feels like compliance overhead right now. In two years it’ll feel like table stakes.
The skills that matter most aren’t AI skills. This sounds counterintuitive in a piece about building tech teams in the AI era, but Korn Ferry’s 2026 data found that 73% of talent acquisition leaders rank critical thinking and problem-solving as their #1 need — AI skills rank fifth. The most valuable hires aren’t AI specialists. They’re domain experts who amplify their expertise through AI. Tools change every few months. Judgment and learning velocity don’t.
The uncomfortable bottom line#
There is no established best practice for building tech teams in 2026. Anyone telling you otherwise is selling something. The industry is figuring this out in real time, and honestly, it’s kind of a mess.
But the direction is clear. Less time evaluating artifacts, more time watching people work. Adoption driven by leadership example, not license purchases. Retention built on growth opportunity, not perks. And throughout all of it, the recognition that AI isn’t just another tool — it changes the fundamental economics of what a small team can accomplish, and that changes everything downstream.
The founders who build durable companies through this transition won’t be the ones with the best AI tools or the most sophisticated governance frameworks. They’ll be the ones who understand these principles clearly enough to keep inventing new responses as the landscape shifts under their feet.
That’s the playbook. Not a fixed set of plays, but the ability to read the field and call your own.