
When they leave, who runs it?
Forward-deployed engineering is a new name for an old question: when the people who built it leave, who takes over?

The read
AI's Value Story Sharpens as Organizations Gain Confidence in Governance, Accountability and Workforce AdoptionKPMG · 24 September 2026
Who manages your AI agents once they start working with each other?
KPMG's latest AI Pulse found that the share of large US organisations actively developing or implementing multi-agent systems climbed to 25%, from only 6% in each of the previous two quarters.
So what? That is a huge increase in just one quarter, and it means agent management becomes a critical topic you must have an answer on. We are no longer creating an individual agent, watching it operate and seeing where it goes wrong. We now have agents working with other agents, which means that when it does go wrong, or sometimes when it goes right, we don't know exactly why or where, or what the impact is, until potentially it's too late.
So before the next agent gets connected to another one, I would want to know who in the business is managing them, and whether that person could explain what happened between two agents yesterday.
The question I would sit with
If two of your agents got something wrong between them today, who would know, and how long would it take to find out?
Alex Collins
Co-founder & COO of RAI Digital, a consulting venture builder · Ex-EY Consulting Partner · Writing on agentic AI, venture building, logistics platforms and transformation leadership.
One honest read every fortnight on what agentic AI is doing to consulting, and what to do about it. For people moving from advice to outcomes, rebuilding a firm, or done buying decks. No hype, no fluff.

Forward-deployed engineering is a new name for an old question: when the people who built it leave, who takes over?

Everyone is counting AI agents, and far fewer organisations can say which ones exist, what each one can access and what it did yesterday.

Nearly every enterprise has stalled an AI project in the last year, and the AI was not the problem. The data was built for a monthly report, not for an agent.