
Why Loïc Guyon wants AI agents to prepare, not approve
Loïc Guyon is founder of AI Jungle, an AI transformation partner for owner-led service firms that builds and runs AI agents and trains teams. Guyon argues that agents should prepare decisions and client-facing work while a person retains approval, drawing on the experience of a two-partner executive search firm.
AI agents can create capacity without sending work autonomously: at Asia-Connect, agents booked 139 qualified meetings in three months, 90 of which were held. Guyon argues that human approval should focus on actions with real reputational risk, or it can become a bottleneck.
What’s one opinion about your industry that most of your peers would disagree with?
AI agents should cut the work behind a decision, not remove the human who approves it. Most of my industry treats full autonomy as the finish line and measures progress by how many steps run without a person. For an owner-led service firm, I think that is the wrong target. The firm’s reputation rides on every email that goes out with a partner’s name on it. The agent should do the preparation, and a person should still say yes before anything leaves the firm.
What convinced you? Tell us about the moment, client or result that changed your mind.
The result I point to is Asia-Connect, a two-partner executive search firm. In three months, our agents booked 139 qualified meetings for them, and 90 of those meetings were held. In a firm with two partners, every message to a candidate or a client carries their name. That volume did not come from an agent acting alone. It came from agents doing the heavy preparation, so that two people could handle far more conversations than they could ever prepare by hand.
What’s the usual way people in your field handle this, and what do you do instead?
The usual approach is to automate a process end to end and treat any human step as friction to remove later. The agent sends, and someone reviews the damage afterwards. I do it the other way. The agent prepares the client follow-up, the post or the payment, and before it goes out, the person who owns it sees the exact content and approves or refuses it. A typical case is a follow-up where the delivery date is uncertain. The agent drafts it, a partner checks the date and says yes. We built Pitstop, a free tool, for exactly that approval step.
Where has this view cost you, or where have you had to admit you were partly wrong?
The cost is that my approach is harder to sell as magic. A system that still asks a partner to say yes looks less impressive than one that claims to run on its own, and some people hear the approval step as a lack of ambition. I accept that. The part I keep questioning is how much to ask of the person approving. If every small action needs a yes, the partner becomes a bottleneck and stops reading carefully. Approval has to be kept for the moments that carry real risk to the firm’s reputation.
What should someone in your field stop doing, and what should they start doing?
Stop measuring an AI project by how many steps run without a person. Stop letting an agent send anything to a client that nobody at the firm has read. Start by listing the actions that carry the firm’s name, such as a client email, a public post or a payment, and keep a human approval on those. Then let the agent take over the work that leads up to them: the research, the drafting and the checking. That is where the hours go, and that is where a small firm gets its time back.
Thank you to Loïc Guyon and the team at AI Jungle for sharing an honest view with Leaders Perception readers.
Have a view your industry needs to hear?
Leaders Perception publishes interviews with business owners and leaders who see their industry differently. Five questions, about ten minutes, no calls. It is free and every completed interview gets published.
