Stepping Off the Dashboard: The Quiet Push Against Metric-Driven Engineering

The Backdrop: AI Amplifies the Ranking Machine

DORA's 2025 State of AI-assisted Software Development report, based on about 5,000 respondents, describes AI as an amplifier. It magnifies the strengths of high-performing organizations and the dysfunctions of struggling ones (InfoQ summary). Organizations with a user-centric focus saw AI's benefits amplified. Organizations without one often saw negative effects on team performance.

In large U.S. tech companies, the culture being amplified is often one of monitoring and ranking. In 2025, performance-based selection became explicit, as LeadDev reported. Microsoft cut under 1% of its roughly 228,000 employees for performance, and later gave low performers five days to choose between severance and a performance plan. Meta moved more than 3,000 people out as "non-regrettable attrition." Amazon kept its long-running system of removing its lowest performers. At Microsoft, AI joined the scorecard: in June 2025, developer-division president Julia Liuson wrote that "using AI is no longer optional," and managers were told to consider employees' AI use in performance reviews (Entrepreneur).

DORA's own 2023 report had warned against turning delivery metrics into goals or using them to compare teams, noting that "fixating on performance metrics can lead to ineffective behaviors" (TechTarget). This article is not about that machine. It is about the people and companies moving the other way.

Three Pushes, and Where People Go

The counter-movement is not one trend. It is three different pressures, each pushing people toward something different. The evidence behind each is uneven, and the table says how uneven.

Push	What people want instead	Where they go	How strong the evidence is
Surveillance crowds out meaningful work	Room to do work that matters, measured at the team level	Companies that don't rank individuals, often startups	The push is measured. Example companies exist. Their number and the flow into them are not measured.
Judged by shifting managers and politics	To be judged by customers	Founding a company	Motive surveys and testimony. No causal data.
Pressure without security	Control over their own footing	Founding a company	The worry is measured and rising. No study links it to founding. Actual job loss does raise founding.

Push 1: Surveillance Crowds Out Meaningful Work

When individuals are ranked on activity, the rational move is the work that produces a safe, legible number. Experiments that might fail become bad personal bets. That is Goodhart's law in practice, and it is the reason DORA warned against using its metrics as targets.

The push is measured. In the American Psychological Association's 2023 Work in America survey, 42% of workers monitored with technology planned to look for a new job within a year, against 23% of those who weren't monitored. Monitored workers were also more likely to feel micromanaged (51% versus 33%). This is workplace monitoring in general, not developer metrics specifically.

The alternative exists. Some companies have built their culture around not measuring individuals. Oxide Computer Company has no performance reviews and pays almost everyone the same base salary ($201,227 in 2024, with 23 employees at the time). Co-founder Bryan Cantrill says "the need to quantify performance in order to justify changes to compensation is at the root of much of what's wrong" (The Pragmatic Engineer). Oxide is an example, not a trend. Nobody counts companies like it, so there is no data on whether their number is growing, or on how many engineers move to them because of monitoring.

Push 2: Judged by Shifting Managers and Politics

Under ranking, your score depends on who your manager is this year, how the reorg went, and how well you argued in calibration. Some engineers stop wanting a better manager. They want customers, not internal politics, to be the judge of their work, and that points toward founding something.

The motives are visible. When engineers explain leaving Big Tech, the reasons include layoffs, fear of a performance plan, growing process and committees, and politics, according to accounts Gergely Orosz collected in 2024. In Gallup's 2024 survey, the top reason business owners and aspiring founders gave was wanting to be their own boss (57% and 60%), with earning more money close behind (54% and 60%).

Autonomy is what people gain. Self-employed people report higher work satisfaction than employees. In German, British, and Swiss panel data, and in a survey of about 16,000 people across 23 countries, that gap is not explained by income or hours. It largely disappears once the analysis accounts for how much independence people have in deciding how to work (Benz & Frey).

What is not visible: data showing that political, shifting evaluation actually caused engineers to found companies. The research measured autonomy, not "being judged by customers." The link is a reasonable inference, not a finding.

Push 3: Pressure Without Security

Ranking cultures combine high pressure with little security. You can be cut at any time, sometimes with five days to choose between severance and a performance plan, while the expectations keep rising. Some people respond by looking for footing they control themselves.

The worry is measured, and rising. In Dice's 2026 Tech Sentiment Report, a survey of 1,159 U.S. tech professionals in November and December 2025, 29% were worried about or expecting layoffs at their current employer, up from 22% in 2024. Only 34% had gone untouched by layoffs, down from 56% in 2023. Job stability rose to the second most common reason for switching employers, up from seventh in 2024. Most of that search for stability points to another employer, not to a startup.

Why founding can still feel more secure. In German panel data, self-employed people report a lower average probability of losing their work within two years (12%) than salaried employees (22%) (Hetschko, SOEP). That is the stability of control: no one can rank you out. The same study finds a catch. When self-employed people do expect to lose their work, their life satisfaction drops more than an employee's does.

What is not visible: a study linking the fear of being cut to founding a company. The closest causal evidence is about actual job loss, and from a different cause. In Norway, men whose employer went bankrupt were 3.7 percentage points (155%) more likely to be running their own business four years later than men at stable firms (Røed & Skogstrøm, IZA). That shows losing a job pushes some people into founding. It does not show that living under the threat of a ranking cut does. Swedish data adds a warning: displaced workers who started firms through labor-market programs had lower survival rates, which suggests some of these are necessity ventures (Nyström, Ratio).

Who founds when the cut comes. Losing a job seems to release capable founders, not only people with no other option. The Norwegian authors read job loss as lowering the opportunity cost of starting a business: there is no longer a paycheck to give up. By their rough estimate, job loss triggered about 56% of men's and 68% of women's moves into entrepreneurship in their sample. Businesses started after a bankruptcy survived at about the same rate as others, with around 75% still active roughly four years later. France offers stronger evidence. When a reform let unemployed workers keep their benefits while starting a business, firm creation rose significantly, and the new firms grew, survived, and performed as well as those started before the reform (Hombert, Schoar, Sraer & Thesmar, Journal of Finance). Many people out of work had viable businesses in them. The risk of giving up a salary had been holding them back. Our reading is that in a ranking culture, a layoff does more than push people out. It can free capable people who would never have quit a salary to try. Neither study measures prior ability directly, and the outcomes are mixed: Norwegian founders who came from bankrupt firms ran smaller companies (7.5 employees on average, against 18.7), and more of them saw their income fall (57% against 46%).

The security founders gain is also narrow. It does not stabilize income: in U.S. data, people who had run their own business for ten years earned a median of 35% less than comparable employees (Hamilton, 2000). Nor does it reliably reduce stress: in one study by Stanford and UC Berkeley researchers, 72% of entrepreneurs reported mental health concerns, against 49% of a comparison group (as summarized by ESADE). Founders trade the threat of being ranked out for the risk that customers don't come. It is the stability of control, not the stability of a paycheck.

AI Lowers the Cost of Leaving

All three pushes existed before AI. What has changed is the cost of acting on them. A small team, or one person, can now build and run a product that once needed a larger team. AI shortens the path from idea to working software, and it can deepen the connection with customers: drafting interview guides, summarizing support conversations, and turning feedback into the next small change.

The amplifier works in both directions. Inside a ranking culture, AI raises the quota. Pointed at customers, it amplifies a customer-facing culture, even a culture of one. The shape of founding is changing accordingly: solo-founded companies rose from 23.7% of new startups in 2019 to 30.5% in 2024 and 36.3% in the first half of 2025 (Carta, as summarized by FourWeekMBA).

What the Numbers Show, and What They Don't

If the counter-movement is real, it should show up in how many companies get started. It does, partly.

Total U.S. business applications hit a record of about 5.67 million in 2025, above the 2023 record (Census Business Formation Statistics, via Founder Reports).
But "high-propensity" applications, the ones likely to become employers, peaked in 2023 at about 1.85 million and were about 1.71 million in 2025.
In the Information sector, they fell. High-propensity applications went from about 24,200 in 2021 to about 19,900 in 2025 (FRED). Many software startups register under professional services instead, so this is not decisive, but it does not support a "tech founding boom."
Solo founders are rising, but venture money is not following them. From 2015 to 2024, solo founders were 25% of startups on Carta but 12% of unicorns (Carta).

The honest summary: the pushes are measured, overall founding is at a record, and the shape of founding is shifting toward smaller teams. Data that ties tech founding specifically to monitoring and ranking does not exist yet. Most engineers who leave a job simply take another one.

Moving Abroad: Mostly About Visas

Some people leave the country, not just the company. That flow is growing, but the data says it is driven mainly by immigration, not by engineering culture.

The flow is real. The share of U.S.-based workers who left their jobs to work abroad rose from 2.7% at the end of 2021 to 6% at the end of 2025. Tech is leading, and since January 2025 more U.S. tech workers have moved to Europe than European tech workers have moved to the U.S., reversing the earlier pattern (Revelio Labs data via CBS News).
It is mostly foreign-born workers. In December 2025, 30% of foreign-born job switchers left the U.S., against under 1% of U.S.-born switchers. H-1B registrations for fiscal year 2027 fell 38.5% from a year earlier (Rest of World).
Much of it is absorbed by Big Tech's own offshore hubs. Big Tech added 32,000 jobs in India in 2025, a three-year high, and Microsoft temporarily moves some staff to Vancouver before bringing them back on L-1 visas. We found no data showing large startups absorbing this flow.
Other reasons are named, but not measured. Analysts point to return-to-office mandates, Europe's investment in AI and cloud, and quality of life such as healthcare and work-life balance. Workers interviewed by Rest of World describe fatigue with shifting rules. None of the data we found cites monitoring or ranking as a reason.

For U.S.-born engineers, then, leaving the country to escape metric-driven management is not visible in the data. For people on visas, culture may matter, but the immigration rules dominate.

Outside the Dashboard, You Still Have to Measure

Rejecting metric-driven ranking is not rejecting measurement. Even when no one is ranking you, you need to know whether customers are satisfied, because marketing, product changes, and revenue all depend on it. You also cannot trust your own sense of progress. In METR's July 2025 study, 16 experienced open-source developers expected AI to make them 24% faster. Afterward they believed it had made them 20% faster. Measured, it made them 19% slower (METR). One small study does not prove AI slows everyone down, but feeling fast is not evidence of being fast.

Three principles keep the new numbers honest:

Measure what customers do, not what they say or what you did.
Own the numbers. Measure to decide what to build next, not to report upward.
Never turn them into personal quotas. Make retention someone's quota and you will eventually see cancellation flows designed to be hard to find.

Stage	Measure	What it tells you
No proof anyone needs it	Customer conversations per week	Whether you are facing customers at all.
No proof anyone needs it	Share of active users who would be "very disappointed" to lose the product (the Sean Ellis test)	An early hint of product-market fit. It asks about feelings, so treat it as a prompt, not proof.
People are using it	Cohort retention after 1 and 3 months	Whether anyone stays. A curve that flattens means a core group finds lasting value.
People are using it	Time from sign-up to first value	Whether your AI-boosted speed reaches the customer, or just adds features.
Revenue is coming in	Net revenue retention (NRR)	Whether customers stay and expand. Above 100%, revenue grows with zero new customers.

Sean Ellis introduced the 40% "very disappointed" threshold around 2009 after comparing nearly 100 startups, and has called it "a bit arbitrary" (Justin Jackson). Use it early, before you have behavior to measure. Avoid anchoring on your own commits and deployments. A founder who treats deployment counts as the scorecard has left the company but brought the manager along.

A worked example (illustrative; the team and numbers are invented). Two engineers leave a large company to build a scheduling and inspection-report tool for small heating and cooling contractors. In their first quarter they keep the scoreboard they were judged by at their old job. In the second, they switch to a customer scoreboard.

Metric	Q1	Q2	What it told them
Deployments	140	60	Activity. Halving it was a choice, not a failure.
Features shipped	38	9	Most Q1 features were never opened by a paying customer.
Trial users still active in week 4	22%	41%	The number that mattered. It moved when they cut, not when they added.
Median time to first completed report	3.5 days	40 minutes	Users had to rebuild their customer list by hand. A spreadsheet import fixed it.

AI made features cheap, so Q1 produced many of them, and the old scoreboard rewarded every one. The customer scoreboard showed that the bottleneck was a 3.5-day setup that no new feature could fix.

Where This Breaks

Conditions That Would Make This the Wrong Reading

The counter-movement may be smaller than it feels. The flows from monitoring to startups are inferred from motives and examples, not counted. Tech-sector founding by the Census's employer-likely measure has not risen.

Not every metric is a ranking tool. In safety-critical work, stability metrics such as change failure rate protect people. They should stay, at the team level. Large companies can also point their scorecards at customers, and DORA's findings suggest those that do get more from AI.

The research has limits. The displacement studies come from Norway and Sweden, not U.S. tech. The wellbeing and income research covers self-employment in general. The international data mixes immigration policy with everything else.

The Dashboard Stays. Its Owner Can Change.

The push against metric-driven engineering is not a push against measurement. It is a push about who does the measuring and what they measure. The companies that refuse to rank individuals still measure their teams. The engineers who leave to found companies trade a manager's dashboard for a customer's verdict, which is harder to argue with and arrives in your bank account.

AI amplifies whichever of these you choose. Inside a ranking culture, it raises the bar you are ranked against. Pointed at customers, it raises what a small team can deliver to them. For engineers deciding where to work, and for leaders deciding what to measure, three questions apply to every number:

Does it describe what a customer did, or what a builder did? Keep builder activity as a health check, never as the goal.
Who owns it, and what decision does it change? If the answer is "someone above me, for a review," it is still the old dashboard.
Would it stay honest if someone's pay depended on it? If not, keep it away from pay and rank.

Organizations that can't answer those questions well will keep losing the people who use AI best. The data can't yet say how many. It can say why they go.

For why per-person activity metrics fail once AI changes the unit of work, see Counting the Wrong Thing. For why AI-driven headcount cuts aren't producing returns, see AI Layoffs Aren't Paying Off.

Sources

Engineering Upskilling

Ship Small, Verify Fast, Ask the Customer

Wherever you work, a customer scorecard only pays off if you can change the product quickly and safely when the numbers tell you to. Our hands-on classes teach the delivery discipline behind that: executable specs, automated quality gates, staging environments, and gated deployment pipelines.

Explore the Engineering Curriculum →