S4 Capital’s Monks lets AI agents run unsupervised for up to 48 hours
S4 Capital’s Monks has begun allowing its AI agents to operate without human supervision for stretches of up to 48 hours Most of the autonomous work involves coding on the agency’s own software, and sizable projects can span several days
Before a run starts, staff write the requirements that the agents build against, and the clarity of those requirements shapes the output. The agents then check their own work, splitting tasks, testing code against the original requirements, and reviewing each other’s output as they go. Wesley ter Haar, chief AI and revenue officer at Monks, said the agents ‘were working for 36 to 48 hours and not making mistakes.’ He added that the team can ‘drive some of that with decent prompting,’ after which the agents keep themselves in check. Clients want to see what was prompted and what reasoning the agents used before approving any output. Some clients request a central registry where they can turn agents on or off, track spending, and rate performance. Monks plans to extend the agentic setup beyond software into social and performance marketing, where agents react to trends, comments, or underperforming posts. In those use cases, engagement data serves as the feedback loop that agents check their work against, similar to how requirements guide coding agents.
“We had coding agents that were working for 36 to 48 hours and not making mistakes,” Wesley ter Haar, chief AI and revenue officer at S4 Capital’s Monks said.
“You can drive some of that with decent prompting,” Wesley ter Haar, chief AI and revenue officer at S4 Capital’s Monks said.
Until recently, developers at Monks had to write out detailed instructions for how agents should divide tasks and pass work between them.
What we know
Some runs go on for 36 to 48 hours and sizable projects can take several days, said Wesley ter Haar, chief AI and revenue officer at S4 Capital’s Monks.
Asked on S4’s half-year earnings call in August what surprised him most about AI this year, he pointed to how long agents could now work unsupervised.
Early next year a client (which Ter Haar did not name) will move their social work onto this rebuilt agentic setup running from insights through to community management.
Until a few months ago developers had to build that scaffolding themselves, writing detailed instructions for how agents should divide tasks and pass work back and forth.