Every conversation I have with startup founders right now hits a strange note. They’re building fast — agents, tools, workflows — and they’re proud of their velocity. They should be. But when I ask what their moat is, the answer keeps getting thinner.
“We move faster.” “We don’t have legacy.” “We can iterate without committees.”
These were real advantages two years ago. They’re evaporating.
The old math: speed as differentiator
The startup playbook for enterprise software was simple: find a workflow that a big company does badly, build a tool that does it well, sell it to the people who live with the pain. The enterprise was too slow, too bureaucratic, too risk-averse to build it themselves. By the time their IT department finished the RFP process, you’d have 50 customers and a Series B.
That math depended on code being expensive. A startup’s advantage was that a small, focused team could produce software faster and cheaper than an enterprise’s internal engineering org. Distribution and data were the enterprise’s advantage, but code was the startup’s.
AI broke that equation.
What actually changed
The cost of producing software collapsed. Not by 20% — by an order of magnitude. Retool’s 2026 Build vs. Buy report found that 35% of enterprises have already replaced at least one SaaS tool with a custom internal build, and 78% expect to build more of their own tools this year.
Retool CEO David Hsu put it bluntly: “Two years ago, a custom internal tool might take an engineering team weeks and cost six figures. Today, a business operations lead with the right platform can have a working prototype in a day or two. That’s a structural change, not a cyclical one.”
When code gets cheap and stays cheap, the startup’s primary advantage — velocity of software production — stops being differentiating. What’s left? Data, distribution, customer relationships, institutional knowledge. All things enterprises already have in abundance.
This isn’t theoretical. Banking — one of the most risk-averse, compliance-heavy industries — is the leading indicator. Deloitte’s 2026 banking outlook documents the shift: banks are moving to hybrid models — buy foundation models, build custom layers on top of proprietary data — specifically because generic SaaS can’t access or leverage what makes their data valuable. JPMorgan spent $18 billion on technology last year and builds most of its AI in-house — 200,000+ employees using their internal LLM Suite daily, scaling from zero to that in eight months.
Meanwhile, Semafor reported this week that “the gap between what a well-funded IT department can offer and what an inspired intern can vibe code over a weekend is closing. Actually, the intern may already have won.”
The intern isn’t at a startup. The intern is inside the enterprise.
The leverage inversion
Here’s the part that most people miss: AI didn’t level the playing field between startups and enterprises. It tilted it toward enterprises — if they choose to act on it.
Consider what an enterprise has that a startup doesn’t:
- Proprietary data at scale. AI systems trained on better data produce better outputs. A startup building a customer analytics tool trains on synthetic or small-sample data. The enterprise has years of actual customer behavior. When code is cheap and data is expensive, the rational move is to build in the data-rich layers.
- Existing distribution. A startup needs to acquire every customer. An enterprise deploying an internal tool has 10,000 users on day one.
- Domain expertise. The people who understand the workflow best already work there. They know the edge cases, the exceptions, the reasons why the “obvious” solution doesn’t work. With AI-assisted development, those domain experts can now build the tool themselves.
- 60% of enterprise builders are already building outside IT oversight — that’s not shadow IT as a problem, that’s a demand signal. The builders are already inside the building.
The Forbes analysis of banking puts it precisely: “Banks are not building more because they suddenly have better engineers. They are building more because AI turned their data from a dormant asset into a working one.”
That sentence applies to every enterprise in every industry.
Why most enterprises haven’t noticed
If the leverage has shifted, why are enterprises still buying? Three reasons:
1. Procurement inertia. The buying machinery is built. Vendor relationships exist. The RFP process is understood. Building requires a different organizational muscle — one that most enterprises haven’t developed.
2. Leadership mental models. Many executives still operate with the mental model that building software is what startups do and buying software is what enterprises do. This was rational for 30 years. It’s increasingly wrong.
3. The pilot trap. Enterprises that try to build often start with ambitious “AI transformation” programs that attempt to boil the ocean. They fail. An MIT study found that 95% of generative AI pilots fail. But the failure isn’t because building doesn’t work — it’s because they’re building the wrong things at the wrong scope.
The companies getting it right aren’t running transformation programs. They’re letting individual teams solve their own pain points with AI-assisted development, then governing the results. Start narrow. Start where the pain is. The tool that replaces one broken workflow is worth more than the platform that promises to replace everything.
What the winners are doing differently
The pattern I see in the organizations that have figured this out:
They ask “what should we own?” instead of “what should we buy?” The new dividing line isn’t capability — it’s data gravity. If owning a capability compounds your proprietary advantage, build it. If a vendor’s scale creates structural advantages (fraud detection across millions of transactions, for example), buy it. The discipline is knowing which is which.
They treat internal builders as a feature, not a bug. That 60% building outside IT oversight? The right response isn’t to lock it down. It’s to give those builders governed environments where they can move fast with real data, proper permissions, and audit trails.
They’ve reclassified AI from “innovation” to “infrastructure.” When JPMorgan moved AI from an innovation line item to core infrastructure — alongside payments and cybersecurity — it signaled that AI had stopped being a science project. That reclassification drives fundamentally different investment and hiring decisions.
They measure time-to-capability, not vendor count. The metric that matters is: how fast can we go from identifying a workflow problem to having a working solution in production? If internal building (with AI) is faster than procurement, the math speaks for itself.
What this means for startups
This isn’t a death sentence for startups — but it does narrow the viable playbook.
Startups that win will be the ones building things enterprises genuinely can’t build themselves: capabilities that require cross-company data (fraud detection across hundreds of institutions), infrastructure that only makes sense at platform scale, or tooling so specialized that no individual enterprise would invest in it.
The startups that lose will be the ones whose moat was “we’re faster than their IT department.” That moat just got drained.
The leadership shift
The real barrier isn’t technical. The technology to build internally is already there and getting cheaper by the month. The barrier is the mental model at the top.
The executive who still thinks “we’re not a software company” is making a category error. Every company that uses software in its operations — which is every company — now has the option to build the specific software it needs, tuned to its specific data and workflows, at a fraction of what it cost even two years ago.
The question isn’t “can we afford to build this?” It’s “can we afford to hand our proprietary data and workflow knowledge to a third party, knowing that every competitor will eventually have access to the same vendor’s tool?”
The answer, increasingly, is no.
The startup advantage was real when code was expensive. AI made code cheap. What’s expensive now is data, domain expertise, and distribution — all things that enterprises already own. The organizations that internalize this shift won’t just match startup velocity. They’ll exceed it — because they’re building on top of assets that took decades to accumulate.
The ones that don’t will keep buying tools from startups that are, increasingly, building on thinner and thinner ice.