The End of the Talent Moat: When Anyone Can Build Software
Executive Summary
For most of my career, the wall between an IT ambition and its execution was never imagination. It was headcount. I have watched more good initiatives die in the resourcing discussion than in any budget review: the product we could specify but could not staff, the integration everyone wanted that sat in the backlog for two years because the three people who could build it were consumed by something else. The talent moat was real, and I spent twenty years on the wrong side of it. Most companies did.
That moat is now draining, and the software market is about to be reshaped by its absence. To understand what is coming, it helps to remember how strange the era of scarcity actually was.
When Owning Talent Meant Owning the Market
The clearest evidence that engineering talent was the industry’s decisive asset is what companies were willing to do to control it — including things that were illegal.
Between 2005 and 2009, Apple, Google, Intel, and Adobe maintained secret agreements not to recruit each other’s engineers. The scheme suppressed mobility and wages for roughly 64,000 technical employees, and it ended in a $415 million class-action settlement — after Judge Lucy Koh rejected an earlier $324.5 million offer as too small against the strength of the evidence. [Source: In re High-Tech Employee Antitrust Litigation, settlement approved 2015] Four of the most sophisticated companies on earth risked antitrust liability rather than compete openly for programmers. That is how much the talent was worth.
Where the companies did compete, they competed by acquisition. Mark Zuckerberg told a Startup School audience in 2010 that Facebook had “not once bought a company for the company itself. We buy companies to get excellent people.” [Source: HuffPost, 2010] The product being purchased in an acqui-hire was shut down as a matter of routine; the engineers were the asset on the term sheet.
And when a company could neither retain nor absorb the talent, it paid the talent to do nothing. In April 2025, reporting revealed that Google DeepMind had placed departing UK researchers under noncompete agreements of up to twelve months — on full pay, barred from working, so that they could not carry what they knew to a competitor. [Source: TechCrunch, April 2025] Consider what that arrangement actually is: a market saying that a brilliant engineer’s idleness is worth a full salary, because scarcity makes even a neutralized asset valuable. Salaried silence was a rational purchase.
None of this was irrational. Korn Ferry’s workforce study projected a global shortage of 85 million skilled workers by 2030, with unrealized revenues in the trillions. [Source: Korn Ferry, 2018] When the constraint on every software ambition is people, owning people-time is owning the market. Every executive who lived through those years learned the same reflex: scope the ambition to the team you can actually hire.
The Wall Is Coming Down
Nick Hodges made an argument in InfoWorld this week that most executives will not read but should: code itself is ceasing to matter. His claim is that programming languages are a temporary artifact of human limitations — that the workflow is compressing from “English, to human-readable code, to machine code” toward simply “English, to working software.” The evaluation standard shifts with it: not “is this good code?” but “does it work?”
Let me concede the obvious objection first, because it is currently true. Most AI-built applications today are not good. They are brittle, they handle edge cases poorly, and anyone who has reviewed one knows the distinctive smell of software that nobody fully understands. If you have dismissed the phenomenon on those grounds, you are describing the present accurately.
But you are describing the present. The relevant question is the direction of travel. Every application built with AI assistance becomes part of the corpus the next generation of models learns from; the tooling that was embarrassing eighteen months ago is merely disappointing today, and the gap closes from one model release to the next. This is the same curve executives watched with machine translation and image generation — years of “amusing toy,” then a short window of “surprisingly usable,” then quiet adoption into ordinary workflows. Waiting for the toy phase to end before planning for the adoption phase means planning after your competitors.
My expectation, stated plainly so it can be judged later: within three years, building rather than buying will be the default for a meaningful share of business software. Not the core systems of record — the departmental layer around them.
What a Post-Scarcity Software Market Looks Like
Which software stops being bought first?
The long tail goes first. Every mid-sized company carries dozens of narrow tools — the approval tracker, the vendor portal, the reporting glue between two systems that do not talk. This layer exists as a purchased product only because building it used to require engineers nobody could spare. It is the same departmental territory that low-code platforms colonized a few years ago, and it will be absorbed the same way, faster. The economics are blunt: a per-seat subscription renews annually forever; a tool a capable analyst builds against your own data costs the afternoon it took to describe it.
Systems of record hold out longest, and should. Your ERP, your core banking platform, your regulatory reporting — these carry audit trails, liability, and decades of encoded edge cases. The buy-versus-build line does not vanish; it migrates toward criticality, exactly as it did in the low-code governance debate. What changes is how much territory sits on the build side of the line.
Where does the talent moat go?
Market power does not evaporate; it relocates. When anyone can build the software, the defensible assets become the things that cannot be generated: ownership of the problem, proprietary data, distribution, and trust. A vendor whose only moat was “we employed the engineers and you did not” is in structural trouble. A vendor who owns a regulatory relationship, a data network, or a decade of your operational history is not.
The same logic applies inside the company. The scarce resource stops being the person who can write the code and becomes the person who knows what is worth building — who understands the process deeply enough to describe it, and has the judgment to know what should not be automated at all. Talent scarcity does not end; it moves up a level.
What about the engineers?
Elite engineering still matters — at the frontier, more than ever; the companies paying twelve months of garden leave are not confused about that. What commoditizes is the middle: the routine implementation work that consumed most enterprise engineering budgets. Hodges points at the uncomfortable consequence — if juniors no longer learn by writing routine code, the pipeline that produces seniors needs rethinking. That is a real institutional problem, and it lands on employers, not just universities.
The Mistakes I Keep Seeing
1. Dismissing AI-built tools because today’s are crude. The executives laughing at a brittle AI-generated app in 2026 are the same ones who laughed at “citizen developers” in 2021 and then discovered two hundred ungoverned Power Automate flows in an audit. The lesson was never that the tools were toys. The lesson was that adoption arrives before governance unless you move first.
2. Renewing the long tail on autopilot. Every SaaS renewal this year deserves one new question: could a competent team now build this against our own data, and would owning it outright serve us better? Asked honestly, the answer will increasingly be yes — and the sprawl you are paying for is already larger than your finance team thinks.
3. Treating this as an IT cost story. The deeper event is market-structural: the barrier that kept your competitors from building what you build is draining away too. Cheaper internal tooling is the small prize. The large prize, and the large threat, is what happens when product ambition is no longer rationed by hiring.
Frequently Asked Questions
Will companies really stop buying software?
Not wholesale, and not soon for anything that carries regulatory or fiduciary weight. The shift starts at the departmental long tail, where the cost of an error is an inconvenience rather than a liability. Watch the renewal decisions on narrow, per-seat tools over the next three years; that is where the line will visibly move.
What happens to software vendors?
Vendors whose value is the software itself face erosion from below. Vendors whose value is the network, the data, the compliance burden they absorb, or the integration surface they maintain are far more durable. Pricing will tell you who is who: per-seat pricing on a commodity workflow is the first thing customers rebuild their way out of.
Is this the end of the developer profession?
No — it is a redistribution. Frontier and infrastructure engineering become more valuable, not less. Routine implementation thins out. The role that grows is closer to architect and editor: specifying systems, reviewing what machines produce, and owning the judgment calls. The profession’s real near-term problem is the junior pipeline, because the traditional apprenticeship — years of writing the routine code — is exactly what is being automated away.
The Friday Journal Entry
A century ago, computation at commercial scale was a profession; “computer” was a job title held by rooms full of people. The spreadsheet did not eliminate accounting. It turned calculation from a scarce skill into a Friday routine, and moved the profession’s value up into judgment — what the numbers mean, and what to do about them.
Software development is beginning the same descent into the ordinary, and the companies that prospered by hoarding the scarce skill will not be the companies that prosper from its abundance. The wall I spent twenty years negotiating around was never made of technology; it was made of people we could not hire. Walls made of scarcity do not survive abundance. The companies that owned talent owned the last market. The ones that own judgment will own the next.