Agentic AI · 13 August 2026
Agentic AI Adoption Is Rising Faster Than Disciplined Measurement
TL;DR
- Agentic AI expectation is far ahead of agentic AI reality: per Thomson Reuters' 2026 Future of Professionals research, around 15% of professionals report current agentic AI use, while 53% are planning or considering it and 77% expect agents to be central to workflows by 2030.
- Meanwhile only around 18% currently collect AI ROI metrics — the experimentation wave is moving much faster than the proof-of-value wave.
- The organisations that win with agents will be the ones that treat measurement and guardrails as part of the deployment, not an afterthought.
- The right question about agents is not 'how many tasks can we hand over?' but 'does this expand human leverage or just relocate effort?'
Part 1
The expectation curve has left the deployment curve behind
Agentic AI — systems that plan and execute multi-step work rather than answering single prompts — is the loudest conversation in the field right now. The numbers behind the noise are more instructive than the noise itself.
Thomson Reuters' 2026 Future of Professionals research puts current agentic AI use at around 15% of professionals, with 53% planning or considering it, and fully 77% expecting agents to be central to their workflows by 2030. Expectation is running five years ahead of implementation maturity. That gap is where an enormous amount of strategic optimism — and budget — is currently sitting.
The same research contains the uncomfortable counterpart: only about 18% currently collect AI ROI metrics at all. Organisations are preparing to hand multi-step work to autonomous systems while lacking the measurement habits to know whether their existing, simpler AI is paying for itself.
Part 2
Why the measurement gap is the dangerous part
With assistant-style AI, weak measurement costs you money quietly: you over- or under-invest, but a human is still in every loop. With agentic AI, weak measurement compounds differently, because agents act. An unmeasured assistant is a missed opportunity; an unmeasured agent is an unaudited employee.
The discipline that agentic deployment demands is not exotic. It is the same operational hygiene organisations already apply to people and processes, extended to agents: defined scope, explicit success metrics, error budgets, escalation paths, and review cadences. What is new is that very few AI programs have built those habits even for their current tools — which is exactly what the 18% figure is telling us.
Part 3
Replacement logic versus redesign logic
Underneath the measurement question sits a design question I put to every leadership audience: do AI agents reduce human agency, or expand it? The answer is a choice, not a prediction.
One path uses agents narrowly to replace effort while leaving the work model unchanged. That path can cut cost, but it does not automatically expand capability — and it tends to produce brittle automation nobody fully owns. The stronger path uses agents to expand leverage: one person handling more complexity, more experimentation, and more output, with good judgement still in the loop. The highest upside appears when organisations redesign roles and decision flows around that leverage, rather than just speeding up old tasks.
The strongest professionals will use AI to create more value, not simply produce output faster. The same is true of organisations.
Part 4
A minimum viable measurement kit for agents
If you are piloting agents this year, the measurement kit does not need to be elaborate. It needs to exist before the agent does.
- Scope: a written definition of what the agent may do, touch, and spend — and what it must escalate.
- Baseline: the current cost, cycle time, and error rate of the process the agent is entering. Without a baseline, every result is an anecdote.
- Success metrics: two or three numbers, owned by a named person, reviewed on a cadence — not a dashboard nobody reads.
- Error budget: how often the agent may fail, in what ways, before it is pulled back for redesign.
- Human leverage check: after ninety days, is the team doing higher-value work, or the same work with extra supervision overhead?
Key takeaways
If you only keep four lines
Adapted from my UNSW Sydney guest keynote 'Six Big Questions Shaping the Future of Analytics in the AI Era', April 2026. Agentic adoption and ROI figures cited from Thomson Reuters' Future of Professionals Report 2026.
About the author
Rushi Vyas — AI keynote speaker, trainer, and consultant.
I help organisations across Australia and Asia-Pacific build practical AI capability through keynotes, AI fluency and literacy programs, consulting, and custom AI products — with $32M in commercial outcomes supported across governments, universities, and brands.
Explore next
Related pages
Related pages that help buyers move from first interest to a clear next step.
Next step
Want this thinking applied to your organisation?
The same material runs as keynotes, executive sessions, and capability programs — tuned to your sector and AI maturity.