The Future of Work: How AI Agents Will Reshape Business Process Discovery in the Next Three Years

The automation field has always been animated by a tension between human judgment and machine execution. Humans identify processes, define scope, and make qualification decisions. Machines execute the automations that humans design. This division of labor has been stable for the past decade, and it's about to change.

The emergence of agentic AI - systems where language models orchestrate complex, multi-step tasks with limited human supervision - is beginning to close the loop. AI agents can now conduct discovery interviews (VARA does this today), generate process maps, produce qualification assessments, and recommend automation investments. In the next three years, these capabilities will extend further: agents that proactively identify automation opportunities from process logs without human prompting, agents that design and test automation solutions, and agents that monitor and optimize deployed automations without developer intervention.

What Agentic AI Means for Process Discovery

Process discovery - the systematic identification and qualification of automation opportunities - has historically required human effort at every stage: designing the intake process, conducting the interviews, reviewing the findings, making the pattern recommendation. AI has been a tool in this workflow (ML models for pattern classification, NLP for information extraction from interview transcripts) but humans have remained in the loop at every decision point.

Agentic systems are changing this. A multi-agent discovery architecture might work like this: a discovery orchestration agent continuously monitors process execution data (from ERP, CRM, ticketing systems) and identifies statistical anomalies - unusual variance in cycle times, unexpected exception rates, rework patterns - that suggest automation opportunity. When it identifies a candidate, it dispatches an intake agent (like VARA) to conduct the discovery interview. The interview output is reviewed by a qualification agent that applies the Problem-First framework, generates the assessment, and routes the qualified candidate to the appropriate development queue.

The human's role in this architecture shifts from process execution (conducting interviews, applying frameworks) to process governance (setting criteria, reviewing edge cases, approving high-value candidates, managing the overall portfolio). This is a fundamentally better use of human expertise - judgment where judgment is scarce, oversight where oversight is necessary.

The Agentic Architecture: Components and Timeline

Now (2026): Conversational AI Intake and Qualification

We are here. AI agents that conduct structured discovery interviews, apply qualification frameworks, produce assessed reports, and recommend automation patterns are production-ready and in enterprise deployment. VARA represents this tier. The human remains in the loop for final qualification approval and priority decisions, but the intake and assessment work is largely automated.

Near-Term (2026–2027): Proactive Opportunity Discovery

The next tier - agents that proactively discover opportunities from operational data rather than waiting for human submission - is in active development at several platforms. Celonis's AI-powered opportunity identification, UiPath's Autopilot, and Microsoft's Copilot for Process Mining all represent steps in this direction. In 12–18 months, expect to see commercially mature tools that continuously scan process execution data and surface qualified automation candidates without human prompting.

Medium-Term (2027–2028): Design and Test Automation

The tier after discovery is design. Agentic systems that can draft automation designs - RPA workflow structures, API integration specifications, BPM workflow maps - from qualified intake reports are in early development. The output won't be production-ready without expert review, but it will represent a significant reduction in the time from qualified candidate to development-ready specification. This tier is likely 12–24 months from commercial maturity.

Longer-Term (2028–2029): Autonomous Optimization

The longer-term vision - agents that monitor deployed automations, identify performance degradation, diagnose root causes, and implement fixes without developer intervention - is technically feasible and actively being prototyped. Commercial maturity for this tier is likely 3–4 years out, with early applications in low-risk, well-monitored environments (data integration, reporting automation) preceding deployment in high-stakes operational contexts.

What This Means for CoE Teams

The progressive deployment of agentic capabilities doesn't eliminate the CoE - it transforms it. The work that consumes the most CoE capacity today (intake interviews, pattern qualification, business case development) will increasingly be handled by AI agents. The work that will grow in importance (governance design, exception handling strategy, agent oversight, business stakeholder relationships, cross-domain portfolio management) requires the specifically human capabilities that AI augments rather than replaces.

CoE leaders who are thinking strategically about the next three years should be asking: what decisions should remain with humans in an agentic automation architecture? Where does human judgment add value that an agent cannot replicate? How do we govern agent-driven discovery and deployment responsibly? These questions are more important than any platform choice or technology decision - they define the organizational design of the automation function for the coming decade.

"The question isn't whether AI agents will transform automation discovery. They already are. The question is whether your organization is designing for the world where agents do the discovery and humans do the governance - or still designing for the world where humans do everything."

- Chief Automation Officer, Global Technology Company

The Near-Term Practical Agenda

For practitioners focused on the next 12 months rather than the next 3 years: the agentic future is most effectively prepared for by building the data and governance infrastructure that agentic systems will rely on. Clean process event logs (for proactive discovery). Structured intake data (for agent-driven qualification). Documented automation patterns and decision criteria (for agent-driven design). Organizations that invest in these foundations now will be best positioned to capture value from the next wave of agentic tooling as it matures into commercial deployment.

Frequently Asked Questions