Frontier
What comes after agents: the orchestrated enterprise
Autonomous functions are the first step. The next is functions that share context and act together. A look at where the technology is heading and what it will require.
Key takeaways
- The path runs from features, to functions, to a coordinated organization.
- Coordination needs shared context, identity, authority and evaluation to be solved at the organizational level.
- Independent forecasts expect agentic AI to become a large share of enterprise software over the next decade.
- Organizations that design for autonomy now will compound an advantage.
The first generation of enterprise AI added assistants to existing software: a button that drafts an email, a panel that summarizes a document. The second generation, which is now arriving, gives AI systems responsibility for outcomes inside a function. A finance system prepares the close. A revenue system turns every enquiry into a qualified conversation.
The third generation is less discussed and, in our view, more consequential. It is what happens when those functions stop operating in isolation.
From features, to functions, to the organization
Each step widens the scope of what an AI system is responsible for:
- Features. AI inside a single application, helping a person do a task faster.
- Functions. AI that owns a defined business function end to end, acting inside several systems, under defined authority.
- The orchestrated enterprise. Functions that share context and hand work to one another, so a signal in one part of the business informs every other.
Consider a large deal being signed. In most organizations, that fact then travels by email and meeting: finance adjusts the forecast, operations plans capacity, people plan for onboarding, customer success prepares a welcome. Each step depends on someone noticing and relaying. In an orchestrated enterprise, each function reacts from the same shared context, and nobody has to relay anything. You can see a small version of this in the interactive example on our homepage.
What independent forecasts say
Analysts expect the shift to be substantial. Gartner projects that by the end of 2026, 40% of enterprise applications will include task-specific AI agents, and that these will evolve toward multi-agent ecosystems by 2029. In its best-case scenario, it projects agentic AI accounting for around 30% of enterprise application software revenue by 2035, more than $450 billion, up from about 2% in 2025. Separately, Gartner expects that by 2029, agentic AI will autonomously resolve 80% of common customer service issues, lowering operating costs by around 30%.
Forecasts are not guarantees, and these are Gartner's scenarios rather than ours. But the direction is consistent across sources: more autonomy, in more functions, operating together.
What orchestration will require
Making functions work together is mostly not a modelling problem. It is an organizational engineering problem, and it has at least four parts.
- Shared context. Systems need a common, governed layer of facts about customers, orders, commitments and constraints, so that they reason from the same picture.
- Identity and authority for software. Every autonomous system will need a verifiable identity, scoped permissions and an accountable owner, in the way an employee has.
- Coordination protocols. When one system's output becomes another's input, there must be clear rules about who decides what, how conflicts are resolved and when people are brought in.
- Continuous assurance. Periodic audits will give way to evaluation in production, and regulators will increasingly expect evidence rather than assurances.
Beyond today's agents
Further out, there are research questions we find genuinely open. How does a system stay on mandate across a thousand steps? Can an organization remember why a decision was made, and not only what was decided? How do you prove a system is ready for more responsibility? These are the questions behind our research agenda.
What to do now
You do not need to wait for the third generation to prepare for it. Start with one function. Define its authority. Measure its outcomes. Build the habits, the records and the controls that will make the second and third functions easier. Organizations that treat autonomy as an organizational capability, and not a series of tools, will find that each step compounds.
Sources
Keep reading
More from the lab
Playbook
How to choose the first function to automate
The first function you give to an autonomous system decides how much trust, budget and momentum you have afterwards. A practical way to choose it.
Thinking
Autonomy is a dial, not a switch
Most conversations about AI ask whether to automate something. The better question is how much authority to grant, and how that authority is earned.