There’s a joke inside Sid Lee these days about getting “Claude-pilled.”
It’s the moment when someone stops thinking of AI as something you talk to and realizes it can actually do work. Before and after. Jean-François Lavigne has watched it happen across the agency: mild disbelief, then conversion.
As Creative Director and Innovation Lead, Lavigne is part developer, part art director, having spent his career moving between code and canvas. His mandate now is to figure out what becomes possible when creative ambition meets the right technology, then make that visible then make that visible to clients, and to the agency itself.
Since January, when tools like Claude Code and Codex began circulating internally, that visibility spread further than he anticipated. Project managers. Salespeople. People who don’t code and never expected to. When they were unsure where to begin, Lavigne’s advice was simple: tell it what you want. Two days later, they came back: “Look what I built.”
“That shift is probably the clearest sign of impact,” he says. “People who would never have touched certain kinds of work suddenly feel able to prototype, test ideas, and make things themselves.”
4:30 on a Friday and Still Going
It’s late on a Friday at Sid Lee, and more than 100 people are still actively using the agency’s internal AI platform. Working, not wrapping up. Lavigne noticed the number almost in passing, glancing at a usage dashboard. To him it was unremarkable — which is the point.
AI at Sid Lee is infrastructure, like Wi-Fi. Every employee has access to a chat environment built on open-source software, with multiple models, personalization, and a growing library of agents. Creative teams have image and video tools. Developers have coding agents, code review, quality checks. A Photoshop plugin gives creatives direct access to image models from inside the tool they already use — AI that disappears into the workflow rather than demanding a new one. The base layer is universal; advanced capability is allocated by practice and need.
The infrastructure is proprietary by design and runs deeper than creative tooling. AI threads through the entire operation, from new business to final delivery. Agents scan procurement and RFP sites, flag relevant opportunities, analyze requirements, and draft response material. Inboxes are monitored and triaged. The building blocks developed for one workflow get recombined for the next. Sid Lee has built its own stack rather than buying one. A vendor’s roadmap would put the agency in reaction mode, not out front.
“Our clients expect originality and velocity,” Lavigne says. “Waiting for integration is not a viable strategy.”
Harder to Unlearn
Widespread use doesn’t mean willing use. Lavigne sees the gap.
The resistance, he believes, is rarely rational. His analogy: don’t follow the Roomba around the house. Some people do exactly that with autonomous agents, watching until the tool fails, almost relieved when it does. The gap between what the tool does today and what it did six months ago is significant, Lavigne says. Running yesterday’s skepticism against today’s capability, he argues, is measuring the wrong thing.
In an agency, thinking and making have traditionally lived in separate rooms. Strategists frame the problem. Account leads manage the client. Creative specialists — writers, designers, 3D artists — execute the vision. The creative director moves between those rooms, translating direction into craft without necessarily doing the craft themselves.
AI knocked down the walls. The strategist can now generate a first draft. The account lead can prototype a concept. The creative director finds themselves not just directing the work but inside it. “It exposes how much of creative work used to depend on that back-and-forth with another human expert,” says Lavigne.
Faster, Until It’s Real
The same gap shows up on the client side, only it arrives dressed as strategy. Leadership comes in with a mandate — use AI, be more efficient — but when actual creative work begins, the tone shifts. Lavigne hears it consistently: worries about quality, whether the work will feel generic, what it signals to use AI at all. There’s cultural baggage too — what it means for artists, for jobs, for the integrity of the idea. The directive says go but the room says wait.
“Clients often come in thinking AI is mainly about cost reduction,” he says. “One of the first things we have to say is, ‘That’s not the most useful way to think about it.’”
A campaign that once required months of specialized production can now be prototyped in days. A background extends, an environment shifts, a variation emerges mid-process. The AI involvement is invisible in the final output. The creative judgment behind it is not.
Hours Were Never the Point
That distinction matters most when the conversation turns to money. AI forces a reckoning with an assumption the industry has long left unexamined, that value correlates to hours.
“Clients are not only paying for hours of mechanical effort,” Lavigne argues. “They are paying for problem framing, taste, judgment, validation, and accountability.” Where AI compresses routine work, the efficiency passes on honestly. If the tool did in minutes what once took days, the invoice should reflect that. At Sid Lee, time gained goes back into the work itself — sharper analysis, stronger creative development, a more rigorous read on what’s actually good.
That requires a more honest accounting of contribution than the industry has historically demanded of itself. Did AI originate the output, or accelerate the process, or function as one tool among many in a largely human-led workflow? The answer changes what goes on the invoice and who can credibly sign off on it.
The billing question is really an authorship question and just one of many Sid Lee is working through. “We have moved away from treating all AI involvement as equivalent,” Lavigne says. “There is a meaningful difference between AI originating an output, AI accelerating a process, and AI functioning as one embedded tool among many in a largely human-led workflow.”