Enterprise software is being rebuilt for AI agents, not humans. Most organisations are not rebuilding how they work at the same pace. That gap is where the value is being lost.
We see the same pattern across organisations of every size.
The technology has moved. The way people work has not.
Investments have been made. Platforms have been built. Pilots have run and shown results. But the teams behind those organisations are still solving problems the same way they were ten years ago, with AI bolted onto the end of a process that was never built for it.
The result is always the same. Pilots that never scale. Slow delivery. Leaders who cannot point to a return.
How enterprise software is being rebuilt for AI agents
For decades, every piece of enterprise software was built around one assumption. A person would always be in the middle.
Someone would log in, read the data, make the decision and move the work to the next step. Entire operating models were built around this. Teams, processes, approval chains, governance structures, all designed with humans at the centre.
That assumption is now being removed from the software layer.
What is replacing it is headless software. Applications rebuilt so that AI agents can operate them directly. No browser. No login. No human in the middle. The application becomes infrastructure. The agent becomes the operator.
In April 2026, three of the world’s largest technology companies confirmed this direction within eight days of each other.
Salesforce announced Headless 360 at TDX 2026, making its platform callable by AI agents through APIs, MCP tools and CLI commands. No browser. No human in the middle. OpenAI launched workspace agents in ChatGPT, designed to run complex workflows across tools like Slack, Salesforce, Google Drive and Notion. Google extended Gemini Enterprise with agent-building and workflow tools aimed at the same shift.
Microsoft had already moved in this direction with Copilot Studio, its low-code platform for building agents that operate workflows and act autonomously across enterprise data and tools. ServiceNow has been rebuilding IT workflows around agents that handle tickets without a human driving every step.
These are not future plans. They are live products. The software layer is changing now.
By 2028, 33% of enterprise software applications will include agentic AI. Gartner, 2024
By 2028, 15% of day-to-day work decisions will be made autonomously by AI agents. Gartner, 2024
The technology is moving faster than the operating models built to use it. That gap is not a technology problem. It is a delivery problem.
What AI-first teams actually look like
There is a big difference between a team that has AI tools and a team that works in an AI-first way. We work with both. The difference in what they produce and how fast is significant.
Legacy teams are large. They write long documents and detailed plans. They review and approve at every stage. Governance happens at the end of the delivery cycle. Coordination runs through people at every step.
AI-first teams are small. They work through working demos, not documents. They try things, learn quickly and adjust. They use accelerators, pre-built components that compress delivery time and improve quality. Governance is built into the workflow from the start, not added on top at the end.
The difference in output is not small. Small AI-first teams consistently deliver working solutions in weeks that larger legacy teams cannot deliver in months. We see this in our own work every time. The constraint is almost never the technology. It is the way the team approaches the problem.
Only 1% of companies describe their AI as mature. Most are still experimenting. McKinsey State of AI, 2024
That number says a lot. After years of significant AI investment, the gap between experimentation and real operational maturity is still enormous. The reason is not the technology. It is that organisations are trying to use AI inside ways of working that were designed before AI existed.
JPMorgan Chase has deployed AI tools to more than 60,000 employees across more than 200 use cases. Walmart is rolling out four AI ‘super agents’ across shopping, staffing, supply and support. Work that used to require significant human coordination. These organisations did not just buy better tools. They redesigned how work happens. That redesign is what creates the return.
Why putting AI into old workflows does not work
The failure pattern is consistent. We recognise it immediately.
An organisation puts AI into an existing workflow. Early results look good. Then the programme stalls before it scales.
The reason is almost always the same. AI was placed inside a process designed for human coordination. Governance still runs through committees. Delivery still flows through approval chains. The bottleneck moved. It did not go away. The technology worked. The process around it did not change.
When the process is rebuilt with agents native to it, the people previously tied up in coordination move to work that needs real expertise. Handling exceptions. Making regulatory judgements. Managing client escalations. Output improves. Value increases. But it requires deliberate redesign, not just a technology change.
You cannot solve an operating model problem with a technology deployment. The redesign has to come first.
The commercial model is breaking down too: per-seat pricing in the age of AI agents
The commercial side of this shift is as important as the operational side, and in some ways more urgent.
Enterprise software has been built on per-seat pricing for thirty years. One vendor. One contract. One charge per user. It is a model that worked well. It is now breaking down.
If one AI agent can do the work of ten or twenty human software users, charging per human seat no longer makes sense. Vendors know this. The fastest-moving ones have already changed.
70% of software vendors are expected to abandon pure per-seat pricing by 2028. IDC, 2026
Software stocks lost nearly $1 trillion in market cap as AI agents disrupted the traditional software model. Reuters, April 2026
Intercom already charges $0.99 per resolved support ticket. Not per seat. Per outcome. Zendesk launched outcome-based AI pricing in 2024. Other vendors are testing credits, tokens and action-based contracts. IDC expects this transition to be largely complete by 2028.
For organisations holding large per-seat enterprise contracts, this is both a risk and an opportunity. The risk is paying for seat volumes that no longer reflect the value being delivered. The opportunity is a window to renegotiate contracts before new pricing norms are set. That window will not stay open long.
This is not an IT procurement decision. It is an executive committee decision.
Agent traffic versus human traffic: how the web itself is changing
There is a wider dimension to this shift that organisations should be tracking now, even if it does not need immediate action.
Automated bot traffic is expected to exceed human web traffic by 2027. Cloudflare, 2026
AI agents visit approximately 1,000 times more web pages than human shoppers. Cloudflare, 2026
If agents are becoming the main navigators of digital environments, browsing, comparing and transacting on behalf of people, then the logic of digital presence built for human attention starts to shift.
Marketing funnels, product pages and sales processes built around human decision-making will need to adapt. The first layer of discovery and filtering will increasingly be done by agents. Humans will apply judgement at the decision point, not throughout the journey.
This is not urgent for most organisations right now. But the direction is clear. The organisations that start thinking about it now will be better positioned than those who react to it later.
What leadership needs to decide
The organisations that navigate this well have one thing in common. Their leadership is willing to engage with operating model change at a structural level, not just sponsor a technology programme.
In practice, this means three things.
- First, an honest look at where human coordination is the main bottleneck. Not where AI could be inserted. Where redesigning the workflow around agents would create the most meaningful change in speed, cost or outcome.
- Second, a delivery model that moves from assessment to working outcomes in weeks, not months. Small teams. Embedded governance. Pre-built components that compress delivery time. A clear business metric to measure against.
- Third, a procurement review that accounts for the pricing shift. Which vendors are already offering agentic capabilities? Which contracts are locked into per-seat models that no longer reflect value? Where is there room to renegotiate before the market sets new norms?
None of this is a technology project. It is an organisational transformation. It needs a partner who understands both the delivery and the operating model change required to make it work.
The software layer is not disappearing. It is being rebuilt. The organisations that thrive will not be the ones that replaced their stack the fastest. They will be the ones that redesigned how they work before the transition forced them to.
Talk to Inovus
At Inovus, we help organisations unlearn pre-AI ways of working and deliver measurable outcomes through AI-first delivery. We identify where agentic AI delivers real value, redesign the workflows around it, embed governance from day one, and move from assessment to working outcomes in weeks rather than years.
If you are working through what this transition means for your operating model or your technology decisions, we are happy to talk.
About the author
Filippo Di Pisa is CTO at Inovus, a data and AI consultancy backed by La Fosse Group, helping organisations unlearn pre-AI ways of working and deliver measurable outcomes through AI-first delivery.
Sources
Gartner, Forecast: Agentic AI in Enterprise Software, 2024.
McKinsey & Company, The State of AI, 2024.
IDC Worldwide Software Pricing and Licensing Model Forecast, 2026.
Cloudflare Connectivity Cloud Insights Report, 2026.
Reuters Markets, April 2026.
Salesforce Trailblazer DX, April 2026.
Google Cloud Next, April 2026.
Intercom, Fin AI Agent pricing model, 2025.
Zendesk, AI Agent outcome-based pricing launch, 2024.
JPMorgan Chase, Annual Report and AI Strategy Disclosure, 2024.