AI will not replace all your software, but it is changing the build-or-buy decision. Where a process is specific to how you compete, AI makes building your own cheaper than it was. Where it is standard, buying remains safer. Boards should judge each renewal on total cost, risk and measured value, not on the claim that software is dead.
This question is now being asked openly by AI investors and technology leaders: will AI tools that write software undermine the business model of software-as-a-service? This article gives the AIPEX view on what a board should do with its own software portfolio.
Why is the build-or-buy question back on the board agenda?
For two decades the default answer was to buy. Building software was slow, expensive and hard to maintain, so most organisations rented standard applications and adapted their processes to fit. AI coding tools have lowered the cost of the first version of an application substantially. That reopens a question many boards considered settled.
But the cost of the first version was never the whole cost. Software has to be secured, supported, updated, audited and eventually replaced. An application written quickly with AI assistance still needs an owner, a maintenance budget and someone who understands it when it fails. The risk is that organisations compare the subscription they can see against a build cost that leaves most of the lifetime cost out.
Which software is most exposed to change?
Not all software is equally exposed. We find it useful to sort the portfolio into four groups:
| Type of software | Likely direction | Board stance |
|---|---|---|
| Systems of record: finance, HR, core customer data | Stays bought; vendors add AI features | Buy, and govern the AI features the vendor switches on |
| Thin workflow tools built on standard processes | Most exposed to replacement by AI | Review at renewal; test whether AI does the job |
| Processes that differentiate you | Building becomes more viable | Consider building, with a named owner and support model |
| Regulated or safety-critical systems | Change slowly, with assurance | Buy or build only with full evidence and controls |
The second group is where savings are most likely. Many organisations pay for dozens of narrow tools that fill a form, route a request or produce a report. Some of that work can now be done by AI inside tools you already license. The fourth group is where caution matters most, because the cost of failure outweighs the saving.
How should a board measure whether a switch pays?
The same way it should measure any AI investment: in money, against a baseline. For each candidate replacement, management should show the current annual cost of the software and the process around it, the full lifetime cost of the alternative including support, security and staff time, the value lever the change is meant to move, and the owner who will sign for the benefit.
Beware two traps. The first is counting hours saved as a return when nobody redeploys the time, a pattern we describe in You Bought the AI Licences. Where Is the Return? The second is ignoring switching cost: migration, retraining, integration and the months when both systems run in parallel.
What risks come with building more in-house?
Building your own moves risks you used to rent back onto your balance sheet. The board should expect management to address:
- Maintenance. Who fixes the application when a dependency changes or the developer leaves?
- Security. Is AI-generated code reviewed and tested to the same standard as any other code?
- Model dependency. Does the new application rely on one AI provider, and what happens if that provider changes terms?
- Shadow software. Are teams building tools outside any approved process, holding company data in places nobody tracks?
- Accountability. For each in-house tool, who is the named owner?
These risks are manageable, but only if the organisation treats in-house AI-built software as a real asset, with an owner, a budget line and a place in the technology register. The most common failure we see is not a bad build. It is a useful tool that a single enthusiast created, that the business came to rely on, and that nobody else understands when it breaks. A light approval route for internal builds, with basic standards for security, testing and documentation, prevents most of this without slowing good ideas down.
What should the board ask at the next software renewal?
- What does this software cost us each year, including the people who run and use it?
- Could AI already inside our licensed tools do this job to an acceptable standard?
- If we built an alternative, what would it cost to own for five years, not just to launch?
- Does this process differentiate us, or is it standard work we should keep buying?
- What AI features has the vendor switched on, and have we approved them?
Where to start
Start with the renewals due in the next twelve months, and apply the four-group view before anyone signs. Our briefing on how to measure AI ROI in money, not hours sets out the five-part measure a CFO will accept, with a board scorecard. The step on the AIPEX roadmap that fits here is Step 3, Plan: the Executive AI Strategy Day (£1,499 + VAT) agrees priorities, named owners and a benefits scorecard your finance team can track.
Frequently asked questions
Will AI replace software-as-a-service?
Not wholesale. Systems of record and regulated systems are likely to stay bought, with vendors adding AI features. Thin workflow tools built on standard processes are most exposed, and building becomes more viable where a process is specific to how you compete.
How should a board judge whether to build software with AI?
On lifetime cost, not launch cost. Compare the current annual cost of the software and process with the full cost of owning an alternative for five years, including support, security, staff time and switching, and name the owner who will sign for the benefit.
What are the risks of building software in-house with AI?
Maintenance when a developer leaves, security of AI-generated code, dependency on one AI provider, shadow tools holding company data, and unclear ownership. For measuring the return, see You Bought the AI Licences. Where Is the Return? →