Frontier
What comes after agents.
Autonomous functions are the first step. Our research is directed at organizations that can sense, decide and act as one coherent system, safely and at the speed of software.
The question is no longer whether software can act. It is how much authority it can be trusted with, and how that trust is earned.
Three horizons
Deployed today. Designed for tomorrow.
Autonomous functions
Production systems that run defined business functions end to end, including revenue, customer, finance, people and operations, with people in command of sensitive decisions.
The orchestrated enterprise
Functions that coordinate with each other: a signal in revenue that shapes finance's forecast, staffing and operations, without anyone relaying it.
Organizational intelligence
New architectures beyond today's agents: systems with long-horizon goals, institutional memory and the ability to improve safely through experience.
Outlook, 2026 to 2035
The road to the autonomous organization.
Independent forecasts point in one direction. Here is what each milestone means, and where our work meets it.
Agents become a feature of enterprise software
40% of enterprise applications will embed task-specific AI agents by the end of 2026, up from less than 5% in 2025.
Gartner, Aug 2025Every software vendor will claim agents. The differentiator shifts from having agents to trusting them.
The shake-out
More than 40% of agentic AI projects are expected to be cancelled by the end of 2027 because of cost, unclear value or weak risk controls.
Gartner, Jun 2025Organizations will favor partners who can prove value early and demonstrate control.
Autonomous decisions at scale
At least 15% of day-to-day work decisions will be made autonomously through agentic AI by 2028, up from none in 2024.
Gartner, Jun 2025Authority has to be explicit: who decides, within what limits, and who answers for the result.
Multi-agent ecosystems, and service without queues
Agents evolve toward multi-agent ecosystems by 2029, and agentic AI will autonomously resolve 80% of common customer service issues, cutting operating costs by 30%.
Gartner, Aug 2025 and Mar 2025Coordination between systems, and fluent conversation on every channel including voice, becomes the standard.
Autonomy as the dominant software model
In its best-case scenario Gartner projects agentic AI at roughly 30% of enterprise application software revenue, more than $450B, up from 2% in 2025.
Gartner, Aug 2025Organizations designed around autonomy compound an advantage that retrofitted ones struggle to match.
Forecasts are Gartner's and reflect its stated scenarios; they are projections, not guarantees.
Six shifts ahead
How we expect the next decade to unfold.
This is our own view, shaped by what we see in production. It guides what we build next.
From features to functions
Agents stop being buttons inside apps and start owning outcomes: a close prepared, a pipeline kept current, a queue cleared.
Identity and authority for software
Every autonomous system will need a verifiable identity, scoped permissions and an accountable owner, as an employee does.
Systems that coordinate
Functions will share context and hand work to one another, so a signal in one part of the business informs every other.
Voice as a default interface
Across Asia the phone and the voice note remain primary. Natural speech makes autonomy reachable for everyone.
Continuous assurance
Periodic audits give way to evaluation in production, and regulators increasingly expect evidence rather than assurances.
Institutional memory
Organizations will remember why decisions were made, not only what was decided, and compound knowledge instead of losing it.
Research agenda
The questions we're working on.
Our lab focuses on the problems that stand between capable models and trustworthy autonomy at enterprise scale.
Long-horizon autonomy
Systems that pursue objectives over days and weeks, not single conversations, while staying aligned with intent.
How does a system stay on mandate across a thousand steps?
Multi-system orchestration
Coordinating many autonomous systems so that functions share context and hand off work cleanly.
What is the right protocol for systems that manage systems?
Institutional memory
Capturing how an organization actually decides, so knowledge compounds instead of walking out the door.
Can an organization remember why, not just what?
Assurance & evaluation
Measuring reliability before and during deployment, so autonomy is extended on evidence rather than optimism.
How do you prove a system is ready for more responsibility?
Governed autonomy
Human-machine governance: the boundaries, escalation paths and controls that keep people in command at scale.
Where should judgement stay human, and how is it enforced?
New interfaces
Voice, vision and documents as first-class inputs, so systems work with information the way organizations actually hold it.
What does work look like when software can see and listen?
These are active areas of exploration. Proven results move into our production systems; everything else stays in the lab until it earns its place.
How we research
Research that ships.
Grounded in production
Every idea is tested against real operations, real data and real constraints, not benchmarks alone.
Measured, not claimed
Autonomy is extended only when evaluation shows it has been earned.
Safety as a capability
Control, traceability and graceful failure are engineering problems we treat as core to the work.
From the lab
Read the thinking behind the systems.
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.
ArticleAgentic AI vs AI agents: what is the difference?
The two terms are used interchangeably, and that causes confusion when you buy or build. A plain-English comparison, with examples and a simple test.
ArticleWhy most agentic AI projects stall, and how to avoid it
Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027. The reasons are predictable, which means they are avoidable.
Design partners
Build the frontier with us.
We work with a small number of ambitious organizations to take autonomy further than it has gone before. If that is you, start the conversation.