AI has not killed software. It has changed who builds it.

For a good part of 2026, the market was pricing enterprise software as though AI was about to make a large part of it redundant.

The logic wasn't hard to follow. If an AI agent can do the work previously performed by ten licensed users, businesses need fewer seats. If software itself becomes significantly easier and cheaper to build, businesses may also question why they're paying large recurring licence fees in the first place.

Software stocks were hit hard as a result.

But over the past few months, that story has started to change. Software stocks have recovered strongly, major vendors are still growing, and enterprise customers haven't suddenly stopped buying software.

What has changed is what they expect from it.

Businesses are asking harder questions about what they're paying for, how much manual work is required to operate their platforms, and where AI can take that work on.

We're seeing exactly that across the digital platforms we work with at GammaDX, including Sitecore, Optimizely, Contentful and Shopify.

Our clients aren't looking to get rid of their technology stack. They're looking at how AI can make those platforms easier and faster to operate, and how their teams can get more out of them without continually adding people and process.

That's a pretty significant change.

The bigger shift is how software gets delivered

For me, this is where a lot of the AI commentary gets it wrong.

The opportunity isn't simply removing software licences or headcount. It's removing the manual work sitting between an idea and getting something live.

Take a fairly standard digital release.

There's the requirement, development, testing, regression testing, deployment, analytics, reporting and ongoing optimisation. Historically, people have been involved at almost every step, including plenty of repetitive work that doesn't necessarily require their judgement.

Agents are starting to take on more of that work.

We've been doing this internally at GammaDX over the past year. Initially, that meant creating individual Claude skills for specific work across Sitecore XM Cloud, Optimizely SaaS and PaaS, analytics and other parts of our delivery process.

We're now moving beyond individual skills and into actual internal products.

We've built a QA agent that can regression test releases, a resource planning tool that reconciles capacity against project schedules, an analytics reporting agent, and an account-based marketing intelligence tool. They're small, purpose-built applications running on Railway and connected into the way our teams already work.

None of them have replaced a person.

What they have replaced is a lot of work that people shouldn't really have been spending their time doing in the first place.

And that means our developers, strategists, project managers and digital specialists can spend more time making decisions and less time moving information between systems.

The platforms are changing too

The same shift is happening within the major DXP and CMS platforms.

SitecoreAI and Optimizely Opal are good examples. Both vendors are moving agentic capability deeper into their platforms rather than treating AI as a chat box sitting alongside the product.

That's important.

I don't think buyers are going to be particularly interested in whether a platform has an "AI feature" for much longer. They'll want to know what it actually does.

Can it create and manage content? Can it automate workflows? Can it identify issues? Can it help optimise an experience? Can it reduce the amount of administration required to run a large digital platform?

And increasingly, the same questions will be asked of agencies and implementation partners.

We're already seeing that shift.

Clients want to understand how we're using AI to deliver their work faster, improve QA, reduce repetitive effort and operate their platforms more efficiently. Increasingly, we're also helping their internal teams work the same way.

That is much more interesting to me than adding an AI chatbot to another website.

The software model will change

None of this means the disruption to software is over.

Seat-based licensing is clearly under pressure. AI makes it possible for fewer people to execute more work, while agents themselves increasingly become users of the software. We're already seeing vendors experiment with usage, consumption, credits and outcome-based pricing alongside traditional licences.

There will also be software products that become difficult to justify. Features that once took months to build can increasingly be created in days. Smaller point solutions will face competition from AI-native products, internal applications and capabilities being absorbed into larger platforms.

But enterprise software has something AI still needs: context.

Customer data, content, workflows, permissions, integrations, business rules and years of organisational knowledge already live inside these platforms. As agents become more capable, access to that context becomes more and more valuable.

So I don't think we're watching the end of software.

We're watching software change from something people primarily operate into something that increasingly does work itself.

For digital agencies, that's a fundamental change too.

The teams that do well out of it won't necessarily be the ones with the most AI presentations, partnerships or features on their website. They'll be the ones that can show where agents are actually working inside their business and their clients' businesses today.

That's the direction we've been taking at GammaDX.

And I suspect we're still very early.