Why role mergers work (and when they don't)
What the DevOps merger got right that AI-driven role mergers are getting wrong.
In October 2025, Google Cloud cut more than a hundred design and UX research positions from its cloud division. Some teams lost half their people. The roles eliminated consisted of people who used behavioral data, surveys, and research to decide how products should work. The engineers and product managers who remained were left to ship without them.
This wasn’t an isolated move of eliminating or reducing whole functions. The Designer Fund’s 2026 survey of over nine hundred designers found that sixty-five percent are now doing work that used to belong to PMs, engineers, or design engineers. Forty percent say their PMs and engineers are doing more design. The boundaries between roles that used to sit side by side with distinct expertise are dissolving, and companies are reading the dissolution as a signal to cut headcount.
The logic behind every one of these mergers is the same. AI made design artifacts (think: wireframes, prototypes, user flows) cheap to produce. A cross-functional peer (say, a PM) with the new tools can stand up a multi-screen prototype in an afternoon. So, the reasoning goes, the design role is expendable.
The open question sitting on the table is: were the artifacts the valuable part of the design role in the first place?
A five-month field experiment with 640 Kenyan entrepreneurs suggests they were not.
Researchers at Harvard Business School and UC Berkeley gave half the group access to a GPT-4-powered business mentor via WhatsApp. The other half received a standard business guide.
The impact was not that AI-assisted entrepreneurs did better. Rather, entrepreneurs who were already high-performing did better. Low performers did nearly ten percent worse with AI assistance than without it.
The divergence came down to judgment. The AI presumably generated more advice, more quickly, than the entrepreneurs could have done before. The difference was that high performers recognized which recommendations fit their context and should be acted on. Low performers followed the output less critically.
That result gives an early indicator on the outcome of the role mergers happening now. Give people a tool that automates the creation of artifacts, and the gap between those who apply judgment and those who don’t widens.
The clearest precedent is DevOps, the most successful lateral role merger in recent tech history. Through the 2010s, operations engineering was absorbed into development. It worked, but it was successful for an important reason.
The tidy version says the automation tooling (Terraform, Kubernetes, CI/CD pipelines) took over time-consuming work, streamlining testing and reducing the need for teams dedicated to QA. The nuance was that QA engineers didn’t just run test suites. They decided what to test, where to set the boundaries, and when a release was ready. Those are skills of judgment, and the merger worked not because those needs disappeared, but because engineers absorbed them into their work.
That distinction matters for what’s happening now. The design-and-product merger assumes the same transfer is underway. AI generates the wireframes. AI stands up the prototype. Half the respondents in the Designer Fund’s survey have pushed AI-generated code to production, and only a fifth of them identify as design engineers. The artifacts are cheap, and so the role looks expendable, just as ops looked expendable once a developer could spin up infrastructure in AWS without them.
But the DevOps merger worked because engineers took on the thinking, not just the tasks. In the late 2010s, Intuit underwent this change: across the company, engineers were now expected to account for testing as part of their development cycle, and I watched the impact firsthand as teams adjusted to the new world order. The shift didn’t just lead to the proliferation of automated unit testing. It generated the practice of conducting test swarms: still a people-driven process, but one that now involved multiple functions and maybe an ecosystem quality representative, not exclusively managed by a team focused on quality.
The design consolidation underway now is moving the tasks without moving the thinking behind them. PMs can now produce design artifacts (wireframes, prototypes, user flows), but few are learning to reason over design judgment: when to break a visual pattern, why a flow that tests well in isolation fails when it ships.
The other reason engineers ultimately took on the judgment part around testing is that the metrics moved too. Quality metrics that used to be held by QA became an engineering mandate, which meant engineers didn’t just inherit the test suites; they inherited accountability for what the test suites measured.
Judgment transfers when the metrics move with the role, and someone is still accountable for them. In the DevOps context, that meant test coverage, defect escape rate, release readiness. The design equivalents — task completion rate, time-to-value, usability error rate — need the same ownership transfer, or they stop being tracked at all.
The same survey confirms what happens when accountability doesn’t follow: hiring managers now want AI fluency alongside a high bar for craft, vision, and storytelling, but few companies have updated performance reviews, team structures, or hiring practices to match. The roles merged, yet the accountability didn’t.
The Kenya study explains why the gap is expensive. Hand the design artifacts to people who lack design judgment and the result isn’t worse designs. It’s more designs, produced faster, shipped with more confidence, wrong in ways that are invisible until the product is in front of users. The high performers get better. Everyone else produces more output of lower quality, faster, and nobody sees the cost until retention breaks six months later.
The first question for any lateral role merger is what portion of the absorbed work is execution versus judgment. Execution is the production of artifacts, the running of processes, the operation of systems. Judgment is the decisions about which artifacts to produce, which processes to change, which things to leave alone. If judgment is part of what’s being absorbed, someone has to be accountable for carrying that thinking, or the organization loses it entirely.
From there, three things can happen.
They collapse the role and distribute the tasks. This is what Google Cloud did with its UX research teams. It’s clean on a spreadsheet. It assumes the remaining people will pick up the judgment along with the work. The Kenya study says on average they won’t (some will, some won’t) and the distance between them will widen under the new tools.
They reduce headcount. Fewer people doing the same work, because the execution portion is now tool-assisted. But the people who stay had better be the ones with the judgment, not the ones with the most tenure or the lowest salary. And there’s a risk I’ve written about before: if automating the easy part increases the speed and concentration of the hard part, you may need more people carrying judgment, not fewer.
They free the judgment that was already there. In many roles, people are so over-subscribed on execution — producing artifacts, running processes, managing tools — that they never had room for the judgment work they were capable of. A designer spending three days a week building screens has less time to think about whether those screens solve the right problem. When AI takes the production, those people don’t become redundant. They become available for the work their job was always supposed to be about.
Before collapsing a role into its neighbor, figure out which of these three you’re actually doing. If automation frees up judgment that was needed but never applied, keep the people: they’re about to do the work the role was always supposed to be about. If it increases the concentration of judgment required, you need more of the right people, not fewer. And if it truly decreases the total work needed, take the headcount reduction. But when you do, make sure the accountability for the remaining judgment transfers with it.
Otis, Clarke, Delecourt, Holtz, and Koning, “The Uneven Impact of Generative AI on Entrepreneurial Performance,” HBS Working Paper No. 24-042, rev. October 2025 (SSRN | HBS). Designer Fund and Foundation Capital, AI in Design 2026 Report (report | summary). Google Cloud design layoffs reported by CNBC, October 2, 2025.