Industry Thesis · 8 min read

How AI Is Changing What Enterprise Software Gets Built

enterprise software - How AI Is Changing What Enterprise Software Gets Built

AI is changing enterprise software by collapsing the cost of building it. Companies now commission narrow, custom systems and agents instead of renting broad platforms. Generic tools lose their moat, while enterprise software that owns data, workflow and outcomes gains one. The question shifts from which vendor to buy to what is worth building.

Why enterprise software was built the way it was

Enterprise software looks the way it does because building was expensive. A vendor spent years and millions of dollars on a product, then spread that cost across thousands of customers. Every customer got the same broad feature set, priced per seat.

That model forced compromises. The product had to serve the average buyer, so it fit nobody exactly. Customers adapted their processes to the tool, hired consultants to configure it, and trained staff to work around its gaps.

None of this was a design choice. It was an economic constraint. When a custom build costs more than a decade of licences, you rent. The whole category of enterprise software, from CRM to project management, rests on that arithmetic.

The thesis here is simple: the arithmetic has changed, so the category is changing with it. For the mechanics of that cost shift, see our look at the economics of software development.

What changed when building got cheap

Capable language models changed the cost of producing working code, integrations and interfaces. A small team with AI tooling can now ship what once needed a department. The scarce input is no longer engineering hours. It is a clear understanding of the problem.

That flips who holds the advantage in enterprise software. Vendors held power because they owned the build capacity. Buyers now hold more of it, because they own the context: their data, their process, their edge cases.

Build versus buy is a different question now

The old rule was to buy unless the function was core to the business. That rule assumed building was slow and risky. When a working internal tool takes weeks rather than quarters, the threshold for building drops sharply.

Take a hypothetical company paying for 200 seats at a per-seat price while only 40 people use more than two features. The other 160 seats subsidise functionality nobody touches. A purpose-built tool for those 40 users, plus an agent for the routine work, can beat the licence on fit alone.

Maintenance is the real test

Cheap building does not mean free ownership. Custom systems need monitoring, updates and someone accountable when they break. Founders who ignore this end up with a graveyard of internal tools. The winners treat each build as a product with an owner.

The new shape of enterprise software

The enterprise software being built now differs from the software being bought a decade ago in four ways. Each follows from the same cause: customisation is no longer a luxury.

  • Narrow over broad. Enterprise software targets one workflow done exceptionally well rather than twenty done adequately.
  • Outcome over interface. The product delivers a finished task, not a dashboard a human must operate.
  • Owned data over rented data. Systems sit on the company’s own records instead of a vendor’s silo.
  • Composed over monolithic. Small components connect through open protocols rather than one suite doing everything.

The table below compares the two models of enterprise software side by side.

Dimension Legacy enterprise software AI-era enterprise software
Scope Broad suite for the average buyer Narrow system for one workflow
Pricing unit Per seat Per outcome or per task
Primary user A trained employee An agent, supervised by a person
Customisation Configuration and consultants Built to fit from the start
Data location Vendor silo Company-owned
Moat Switching costs Proprietary data and workflow depth

Agents become the primary user

The biggest shift is who the software is for. For forty years, enterprise software assumed a human at a screen. Increasingly, the user is an agent that reads, decides and acts across systems, with a person reviewing exceptions.

This changes what good enterprise software looks like. An agent does not need a polished interface. It needs clean APIs, predictable responses and permissions it can be trusted with. Mechanisms like tool use and function calling let a model operate software directly, which makes the quality of a tool’s interface a machine-readable question.

Enterprise software that is hard for an agent to operate becomes a liability. A product with no API, locked exports or a UI-only workflow forces a human back into the loop. Buyers will notice, and they will route work around it.

  • Expose every important action through an interface an agent can call.
  • Return structured, consistent data rather than formatted pages.
  • Log every action so a person can audit what the agent did.

The labour consequences are covered in our piece on the talent market. Fewer people operate software, and more people supervise it.

What happens to the big platforms

Large platforms will not disappear. They will change what they sell. The durable value in a platform is its data gravity, its compliance posture and its integrations. The fragile value is its feature breadth.

Expect pressure on three layers of the enterprise software market.

Thin workflow tools

Products that wrap a form, a database and a notification are the most exposed. A company can now build that in-house faster than it can complete a procurement cycle. These vendors must find a deeper moat or lose to a custom build.

Seat-priced suites

When agents do the work, headcount stops predicting usage. Per-seat pricing breaks because fewer humans touch the product while it does more work. Vendors are being pushed toward pricing on outcomes, which is a different business with different margins. The same logic reshapes professional fees, as we argue in the pricing power of knowledge businesses.

Systems of record

The core databases that hold customer, financial and operational truth are the safest layer. Their value grows as agents need reliable data to act on. The risk for them is being demoted from the front door to a back-end store.

What a founder should do with this

This is a thesis essay, not a playbook, but the strategic implications are concrete. A founder running a company between roughly $1M and $50M in revenue has more room to move than a large enterprise, because there is less legacy to unwind.

  • Audit your stack for seat waste. Find tools where usage is concentrated in a few people and the rest is shelfware.
  • Identify the workflows that are unique to you. These are the candidates for custom enterprise software, because no vendor serves them well.
  • Keep your data portable. Avoid contracts that make your own records expensive to extract.
  • Assign an owner to every build. No system ships without a named person accountable for it.

Governance matters as agents gain authority. Decide in writing what your agents may do unsupervised, which systems they may write to, and who reviews their exceptions. A risk framework such as the NIST AI Risk Management Framework is a sensible reference, but the decision about authority belongs to you.

If you sell enterprise software rather than buy it, the lesson is sharper. Your customers can now build a rough version of your product. Your defence is data, depth and trust, not feature count.

Where enterprise software goes next

The direction is clear even if the pace is not. More enterprise software will be built to order, operated by agents and priced by result. Investors are already adjusting how they judge these businesses, as we explore in how investors evaluate growth businesses.

The companies that benefit most will not be those with the biggest software budget. They will be those that understand their own processes well enough to encode them. Knowing your workflow becomes the asset.

Spend on enterprise software also stops being a fixed line item. It becomes an investment portfolio of small builds, each judged on the work it removes. That rewards operators who measure well and cut fast.

If you want to work out what to build and what to keep renting, talk to Studio Máté about building it for you.

Frequently asked questions

Will AI replace enterprise software vendors?

No, but it will change what they sell. Vendors with proprietary data, deep integrations and strong compliance keep their position. Vendors selling thin workflow features or per-seat access to commodity functions face real pressure from custom builds.

Should a mid-sized company build its own enterprise software?

Build where your workflow is unique and no vendor fits well. Keep buying where the function is standard, such as payroll or core accounting. Choose the threshold yourself, and only build what you can name an owner for.

How does AI change enterprise software pricing?

Per-seat pricing weakens because agents do work that humans once did, so headcount stops tracking value. Expect more pricing based on tasks completed or outcomes delivered, which shifts risk from the buyer to the vendor.

What makes software easy for an AI agent to use?

Clean APIs, structured and consistent responses, clear permissions and full action logs. Products that rely on a human clicking through a screen are harder for agents to operate and tend to get routed around.

← Back to all articles