NORBERT BIEDRZYCKI

From Copilots to AI Agents: The Next Software Investment Revolution

From Copilots to AI Agents: The Next Software Investment Revolution

For twenty years, software has been sold by the seat. AI agents do not need seats. That simple fact may be the biggest threat to software valuations since the move to the cloud.

The language of AI has changed again. In 2023 the conversation was about chatbots. In 2024 it was about copilots, assistants that help a human do their job faster. In 2025 the word on every slide is agents: systems that do not just answer questions but execute tasks, call other software, coordinate workflows and complete processes with limited human supervision.

The infrastructure for this is maturing quickly. Anthropic’s Model Context Protocol, released at the end of last year, has become a widely adopted standard for connecting models to tools and data, with OpenAI and others announcing support this spring. Every major enterprise software vendor now has an agent strategy. Salesforce launched Agentforce last autumn with pricing of around $2 per conversation. Customer service platforms are moving to pricing per resolved ticket rather than per agent seat.

Why this matters for software economics

The SaaS business model is built on a simple unit: the user. Revenue grows when customers hire more people who need licences. Public software valuations over the past fifteen years have been built on the predictability of that model: recurring revenue, high gross margins and net revenue retention driven by seat expansion.

Agents break this logic in two ways.

First, if an agent can do work that previously required several people, the customer needs fewer seats. A vendor that charges per user is effectively penalised for making its product more productive.

Second, agents shift value from the application to the workflow. If an agent can operate across multiple systems, the interface of each individual application becomes less important. The system of record still matters, but the layer where work actually gets done may move elsewhere.

New pricing, new risks

Vendors are experimenting with alternatives: pricing per task, per transaction, per outcome. These models can align price with value and potentially expand the market, because software can now compete for labour budgets, not just IT budgets. That is a much larger pool of spending.

But outcome based pricing also brings revenue volatility, harder forecasting and more complex contracts. And it exposes vendors to the reliability problem. If you charge per resolved case, you own the consequences of every unresolved one.

Reliability remains the central weakness. Gartner predicted in June that over 40 percent of agentic AI projects will be cancelled by the end of 2027 because of cost, unclear value or inadequate risk controls. Klarna, which became the poster child for AI customer service last year, said this spring that it would again make it easier for customers to reach a human, after acknowledging that quality had suffered. Security, auditability and accountability for agent actions are still immature.

What I tell boards and investors

When I look at software companies in a portfolio today, I ask three questions. How exposed is the revenue to seat count in functions where agents can do the work? Does the company own the data and system of record that agents will need to access, or only the interface? And is management already experimenting with pricing models that capture value from automation rather than resisting it?

The companies that answer these questions well could emerge stronger. Those that do not may find that their growth slows long before their revenue visibly declines.

Investment question: Will AI agents create a new software category, or fundamentally undermine the economics of traditional SaaS?

Read next

  1. The Physical AI Opportunity: Investing Beyond Software

    The Business of AI series, written in 2026

  2. OpenAI and the consultants: the last mile of AI is organizationalLinkedIn

    When OpenAI announced partnerships with McKinsey, BCG, Accenture and Capgemini, it didn't feel like a breakthrough moment. At first glance, it looked like another predictable move: big tech teaming up with big consulting.

  3. Solow's déjà vu: AI everywhere except in the productivity statisticsLinkedIn

    I've been participating in technology revolutions since the early days of my career, and in AI since the moment it became deployable at scale in real commercial environments. And yet, I increasingly feel a sense of déjà vu.