A 40-engineer fintech I spoke to last month was paying $960/month for GitHub Copilot Business across their team, but their legal department had just blocked it after a compliance review flagged code-completion telemetry flowing through Microsoft's cloud. Their CTO asked me a deceptively simple question: "Can we self-host something equivalent?"

The answer is "yes, but only if you pass three filters." Stack Overflow's 2024 Developer Survey found 76% of developers use or plan to use AI tools, but adoption in regulated industries lags by 20-30 points. The gap isn't skepticism — it's infrastructure. Most engineering teams want private inference but underestimate what "self-hosted" actually costs in GPU capex, SRE time, and model-quality compromise.

This is the decision framework we hand teams considering the switch: when self-hosted LLMs beat the cloud, when they don't, and the three breakpoints that tip the math.

Cursor raised $900M at a $9B valuation in August 2024. Windsurf (formerly Codeium) sold to OpenAI for $3B in 2025. Sourcegraph Cody pivoted to full IDE. Three AI-native IDEs are now mature enough that picking between them is a real question — not "which one works" but "which fits your team's constraints on privacy, latency, and context depth". Stack Overflow's 2025 Developer Survey reported that 62% of professional developers now use an AI coding tool daily, up from 44% in 2024. The same survey showed the choice between tools matters more than the choice of editor: developer satisfaction swings ~20 points depending on which AI assistant, vs ~5 points for underlying editor.

This isn't a "which is best" verdict — it's a decision framework with numbers. We're going to be specific about where each one wins, where each one loses, and where our own IDE heartbeat data from teams running them in production (n=47 teams, ~340 developers) lines up with or contradicts the marketing claims.

A finance lead at a 12-person engineering team opens the cost dashboard. Total monthly burn: $58,000. Four full-time engineers on monthly salary, five contractors on hourly rates, three more outsourced through a vendor invoiced monthly. The dashboard shows a single average cost per developer. It is the wrong number, and every per-feature decision built on top of it is also wrong.

The conventional fix is the 160-hour conversion: divide a monthly rate by 160 to get an hourly equivalent, then compare. The US Bureau of Labor Statistics tracks average annual hours actually worked per employee at 1,791 hours. That is 149 hours per month, not 160. In Kazakhstan, a statutory 24-day vacation entitlement plus 13 paid holidays brings the effective figure closer to 144 hours. The 160 number is a hand-me-down from a country that no longer matches its own data.

Copilot wrote 420 tests for your payments module in two days. Coverage went from 58% to 84%. Release confidence? Unchanged, maybe worse. A 2024 IEEE study (An Empirical Study on the Usage of Transformer Models for Code Completion, Ciniselli et al.) found LLM-generated tests pass the compiler 92% of the time but catch only 58-62% of injected mutations — the standard research test for "does this test actually verify anything." Human-written tests in the same study scored 78%. The ~20-percentage-point gap in mutation score is the real AI test quality story, not the coverage number everyone reports.

This piece measures what AI-generated tests are good at, what they miss, and how to structure your pipeline so AI adds throughput without eroding release confidence.

A senior backend engineer in Almaty earns $5,000/month gross. A CFO scoping a new project does the obvious math: $5,000 ÷ 160 = $31.25/hour. That number lands in a spreadsheet, then in a board deck, then in a quote sent to a customer.

The real cost of that engineer's hour, after overhead, is closer to $46/hour. That's a 48% gap. The 2024 DORA State of DevOps Report puts non-coding overhead at 35–55% of engineering payroll across high-performing organizations. McKinsey's Developer Velocity Index (2023) lands in the same range. Most companies never multiply through. They quote, scope, and forecast on the naive number, then wonder why the books don't close.

The AI coding tool market fragmented into four serious contenders by early 2026: GitHub Copilot, Cursor, Claude Code (Anthropic CLI), and ChatGPT with Code Interpreter. Marketing decks from all four claim "40% productivity boost" — the number is identical, and it's meaningless without measurement. We pulled IDE heartbeat and session data from 112 engineers across 14 B2B teams in Q1 2026 to see what actually saves time.

The punchline: Claude Code users ship 54 minutes of saved time per day; Copilot users ship 28. But the distribution is not what marketing implies — the best tool depends on the kind of work, not the team's "AI maturity".

AI code review sits at the crest of the hype cycle. GitHub Copilot, CodeRabbit, Qodo, Graphite, and half a dozen startups are pitching a future where LLMs catch bugs faster than humans. Microsoft Research and Bacchelli's seminal 2013 study on code review established the baseline we've been measuring against for a decade: human review catches ~14% of functional defects but 68% of maintainability issues. The question now is: does layering an LLM on top actually move either number?

We pulled review data from 100 B2B teams between Q1 2025 and Q1 2026: a mix of teams using AI review, teams not, and teams running hybrid. The pattern isn't what the vendors claim.

Most non-technical CEOs I've met treat engineering as either a black box or a theater. Black-box CEOs ask "how's engineering?" at the executive meeting, accept "we're on track" as an answer, and act surprised four quarters later when the senior architect resigns and the product roadmap stalls. Theater CEOs become amateur engineering managers — they learn to recite DORA metrics, mispronounce "Kubernetes," and inadvertently turn every roadmap discussion into a technical argument they can't follow.

Neither failure mode is about intelligence. It's about the absence of a short, non-technical vocabulary for engineering health. First Round's 2023 State of Startups survey found 68% of first-time CEOs rate themselves "somewhat" or "very" dependent on their CTO for all engineering judgment calls — which is fine until the CTO leaves or disagrees with the board on direction.

This guide is the minimum CEO vocabulary: 6 questions that let you test whether engineering is healthy without pretending to be technical.

An Engineering Director who led a 50-person org well is usually the wrong person to lead a 500-person org well. Not because they lack talent — because the role at 500 is a different job, not the same job at higher intensity. Research from First Round Review's survey of 300+ engineering leaders consistently finds that the transitions at ~80, ~150, and ~300 engineers are where the most senior leader burnouts and quiet departures cluster.

This is a data-grounded guide to the four transitions an Engineering Director faces as the org grows from 50 to 500 — what to let go of, what to pick up, and what our IDE heartbeat data says about the warning signs of a Director who didn't make the shift.

Your best senior engineer just got promoted to "lead." Nobody wrote down whether that means Tech Lead or Engineering Manager, so now she does both. She's reviewing every PR, running every 1:1, planning every sprint, and still expected to ship her own code. Three months in, her output collapsed and so did team delivery. A 2024 Stack Overflow Developer Survey found that engineers in hybrid "lead" roles report 1.6× higher burnout than those on either a pure IC or pure management path. Merging the roles is the single most common — and most expensive — leadership mistake we see.

Tech Lead and Engineering Manager are different jobs with different success metrics, different time allocations, and different failure modes. Pick one per person, or pick both and hire two people.