Monks Runs Unsupervised AI Coding Agents for 48-Hour Stretches
The Development
S4 Capital's Monks is operating AI coding agents in unsupervised runs lasting 36 to 48 hours, with larger projects extending across several days. Wesley ter Haar, Chief AI and Revenue Officer at S4 Capital, confirmed the agents work against a set of requirements written by staff before each run — and then operate entirely without oversight. On multi-day projects, multiple agents split the workload, test their own code against the original brief, and cross-review each other's output autonomously. Critically, ter Haar noted that until a few months ago, this kind of multi-agent coordination required developers to manually write the orchestration scaffolding — the instructions governing how agents divide and hand off tasks. That overhead has now collapsed.
Our Take
The 36-to-48-hour unsupervised run is the tell. This is not AI-assisted development where a human reviews each commit — it is a closed loop running at production quality with no babysitting required. What Monks has demonstrated is that the coordination layer between agents has matured fast enough to transfer meaningful categories of knowledge-work without continuous human intervention. That collapse of the orchestration overhead changes the unit economics of agency services in ways that are not yet priced into client contracts or competitor operating models. Monks is running this on internal software development first, which is low-risk — but the same architecture applies to campaign asset production, data pipeline management, and performance reporting at scale.
What Changed
Multi-agent orchestration — where agents autonomously divide work, self-test, and peer-review output over multi-day cycles — has become a deployable production capability rather than a custom engineering project. The scaffolding that previously required bespoke developer work is now sufficiently mature to remove as a prerequisite.
Marketing Impact
Marketing operations and agency production teams are the immediate target. Work that previously required a developer on standby — internal tooling builds, data pipeline maintenance, reporting automation — can now run as a background process, freeing engineering capacity for higher-order problems.
Competitive Implication
Agencies with the engineering maturity to deploy multi-agent pipelines at production scale will structurally compress their delivery costs ahead of competitors still running human-supervised workflows. Holding companies that cannot match this infrastructure will face margin pressure when Monks-style economics start appearing in pitches.
Strategic Outlook
Monks will extend this architecture beyond internal coding into client-facing production work over the next two quarters. As that happens, per-deliverable pricing models become unsustainable for agencies still billing on hours — and retainer structures will need to reprice around outcomes, not effort.
The Exploit
Action Item
Heads of marketing operations at mid-to-large brands should audit which agency partners have live multi-agent infrastructure now — and use Q1 2027 contract renewals to demand outcome-based pricing that reflects the collapsed cost of AI-driven delivery.