- Is it really worth letting AI agents run unsupervised in your company?
- What did Monks actually do and with what concrete results?
- How does the team structure change when agents work autonomously?
- How does it work in practice for an Italian company: three scenarios compared?
- How to unify the AI budget with the personnel budget: what Monks does and what you can do?
- What AI tools and models can you use to replicate this approach?
- What risks and rules do you need to manage before letting an agent run alone?
- Why are people talking about this now and not a year from now?
- How to get started concretely: the steps to introduce autonomous agents in a company?
- What mistakes do companies make when approaching AI agents for the first time?
- How to tell if the model is working for your business?
- So, is it worth unifying budgets and teams on the Monks model for your company?
- How do we see it at SHM?
- Frequently Asked Questions
- Monks (S4 Capital) runs AI agents for 36-48 hours without supervision on coding, social, and performance marketing, with plans to roll them out to clients.
- The budget for AI tokens has entered the 'talent budget': it is no longer a separate IT item, but competes with salaries for the same resources.
- We need a registry of agents, governance rules, and human figures who select and guide the output, not just approve it.
- In Italy, only 8% of large companies are already experimenting with Agentic Automation, but 25% plan to do so soon: those who start first build a real advantage.
An AI agent working 36-48 hours straight without anyone watching it: it's not sci-fi, it's already happened in production. The question for your company isn't "when will it arrive", but "how do I set things up now that it's already here".
Is it really worth letting AI agents run unsupervised in your company?
It depends on how measurable the work is and how well it is guided by clear criteria. Where precise requirements and objective feedback loops exist, autonomous agents cut costs and save time; where human judgment is still irreplaceable, supervision remains necessary.
Monks, S4 Capital's operating division, ran AI agents for 36-48 consecutive hours on coding tasks without anyone monitoring them. The result: no major errors, much higher speed than human work. The news, reported by Digiday , poses a concrete question: what activities in your company have the same characteristics?
The answer depends on three factors: how precise your requirements can be, the presence of a feedback loop (meaning an objective signal that tells the agent if it's doing well), and your risk tolerance in case of an error.
| Your situation | Agent autonomy | why |
|---|---|---|
| I work with strict rules and checkable output (e.g., code, reports) | Yes | The agent can run for hours or days; requirements act as guardrails |
| Work with real-time measurable KPIs (e.g. social, ads) | Yes | Performance data is the feedback loop that guides the agent |
| Creative work with qualitative judgment (e.g. copy, brand) | Partial | The agent generates options, but an expert is needed to select |
| Work with high reputational or legal impact | No, for now | The risk of unsupervised error is too high |
What did Monks actually do and with what concrete results?
Monks shifted first coding, then social, and performance marketing to autonomous agents that coordinate with each other, test their own work, and correct themselves without human intervention. Financial numbers confirm that the model works.
The starting point was code: precise, verifiable, with requirements that can be clearly written down. The agents receive the specs, divide the work, test the produced code against the original requirements, and check each other. Until a few months ago, this coordination required hand-written instructions from developers. Today, the models do it on their own.
The next step is social and performance marketing. Here, the agents wait for an event — a trend taking off, a wave of comments, a post losing traction — and take action. Engagement data replaces written requirements as a control signal.
S4 Capital's financial data tell the same story from another angle. The 2025 annual report shows a net revenue of £673 million, down 10.8%, but with an operating margin up to 12.1% (+70 basis points compared to 2024): less revenue, more efficiency. Labor costs dropped because agents now do part of the work.
On the creative front, the platform Monks.Flow produced 460 personalized assets for 20 use cases in a single holiday campaign , with 13% lower cost per acquisition and 62% higher conversion than the control.
How does the team structure change when agents work autonomously?
Work doesn't disappear: it shifts. People stop performing repetitive tasks and become responsible for writing requirements, selecting the best output, and deciding the direction. It's a change of role, not of headcount — at least in the initial phase.
Monks is facing an unexpected problem: the agents churn out 50 ideas in 10 minutes, all fitting the brief. Asking for something bolder generates another 50, but only five are actually good. Finding those five takes time and skill. The creative team's job is no longer to come up with ideas: it's spotting the ones worth keeping.
The same goes for work paces. Agile sprints — two-week cycles with daily meetings — have compressed down to two days. The risk is that people stop thinking and just validate. Monks is considering reintroducing deliberate friction to force the team to actually process things, instead of delegating everything to the agent.
- 1 · Strategist / PM Writes the requirements Define objectives, constraints, and quality criteria before the agent starts. The quality of the prompt determines the quality of the output.
- 2 · AI agent Executes and self-monitors Works autonomously for hours or days, tests its own output against requirements, coordinates other agents if necessary.
- 3 · Subject matter expert Select and guide Choose from the options produced, indicate directions for further investigation, identify errors that the requirements did not foresee.
- 4 · Governance team / Tech council Monitor costs and quality Keeps track of active agents, monitors token spending, assigns the right models to the right tasks.
- 5 · Management Decides and responds Maintains final responsibility: agent autonomy does not mean absence of accountability.
How does it work in practice for an Italian company: three scenarios compared?
An e-commerce store, a communication agency, and a manufacturing company all have different needs, but all three can apply the 'unified talent budget' logic starting from the most structured tasks.
The three scenarios below are illustrative: the numbers are there to show the calculation logic, not data from real customers.
| Company type | Activity delegable to the agent | Estimated savings (hours/month) | Main risk |
|---|---|---|---|
| E-commerce (20 people) | Answering comments, optimizing product pages, ads reports | 40-60 hours | Tone not aligned with the brand if requirements are vague |
| Communication agency (8 people) | Copy drafts, A/B variants, creative briefs from client briefs | 30-50 hours | Output selection requires a dedicated senior |
| Manufacturing company (50 people) | Internal reports, tech documentation updates, data analysis | 60-80 hours | Sensitive data: an enterprise contract with the AI provider is required |
The critical variable in all three cases is the same: the quality of the requirements written before the agent starts. An agent with vague instructions produces vague output. Investing time in the setup phase — what Monks calls 'decent prompting' — is the highest leverage you have.
How to unify the AI budget with the personnel budget: what Monks does and what you can do?
Monks stopped treating AI tokens as an IT expense and moved them into the 'talent budget', the same container used for salaries. The logic is simple: if an agent does the work previously done by a person, it competes for the same resources.
The paradigm shift is this: until recently, the cost of AI tools sat in software or IT budgets. Now that agents replace hours of human labor, keeping them in a separate line item skews financial tracking. Monks moved them into the talent budget — the budget earmarked for people — because that's where it truly competes.
The practical problem is that teams are asking for more and more tokens (the processing units that AI models consume for each task). Monks is measuring the demand for 2027 and has created an internal 'tech council' that assigns the most expensive models only to tasks that truly require them. For everything else, it uses cheaper models. This is exactly the kind of optimization that you can replicate even with small teams, as explained in the article on how to cut token costs by picking the right model for every task .
| Cost item | Where it was before | Where it goes now (Monks model) | why |
|---|---|---|---|
| AI tokens (processing) | IT / software budget | Talent budget | Competes with wages for the same tasks |
| Premium AI models | Centralized IT purchasing | Assigned by task from the tech council | Avoid waste: not every task requires the best model |
| AI training | HR training budget | Talent budget or operating budget | It is part of the cost of keeping people productive |
| Agent supervision | It didn't exist | Talent budget (senior hours) | The time spent selecting output is a real cost |
For an Italian SME, the practical step is easier than it seems: just add a line to your operating cost plan comparing the monthly cost of tokens with the work hours they replace. If the ratio is favorable, the budget justifies itself. If not, you are using the wrong model or the wrong task.
What AI tools and models can you use to replicate this approach?
There is no single platform: the choice depends on the type of task, the budget, and the level of control you want to keep. The key distinction is between 'always-on' models and 'event-driven' agents that only kick into action when something happens.
| Agent type | When to use it | Task examples | Relative cost |
|---|---|---|---|
| Continuous run agent | Long and structured tasks (coding, reports, documentation) | Software development, data analysis, catalog update | High (many tokens per run) |
| Event-driven agent | Responds to external triggers (trends, comments, out-of-threshold KPIs) | Social monitoring, ads optimization, customer care | Medium (activates only when needed) |
| Review agent | Checks the output of other agents | Code QA, fact-checking, brand consistency | Low (short and targeted tasks) |
| Lightweight model (low-cost) | Simple and repetitive tasks | Classification, tagging, brief summaries | Very low |
The logic behind Monks' tech council — matching the right model to the right task — can be applied even without a dedicated team. All it takes is a simple matrix: for every process you want to automate, ask how critical an error would be and how complex the required reasoning is. Only high-complexity, high-risk tasks deserve the most powerful and expensive models.
To understand how models are evolving in this direction, it's worth reading the analysis on what changes for SMEs with the new Gemini 2026 models and the one on how headless AI agents are evolving according to OpenAI .
What risks and rules do you need to manage before letting an agent run alone?
Monks clients are already asking to know which models are working for them, where they are hosted, and if there's an enterprise contract behind it. Agent governance isn't a technical detail: it's a contractual condition.
When a Monks client asked about the agents working for them, the questions were precise: which models, where do they come from, is there an MSA (Master Service Agreement, meaning a framework contract) covering them? No one wants unknown models processing company data without a clear contractual basis.
The issue of agent governance is closely linked to content published without supervision: a risk we analyzed in detail in the article on AI agents and images published without consent . The basic rule is: the more autonomy you give the agent, the more robust the control system you build around it must be.
| Risk | Probability | Impact | How to mitigate |
|---|---|---|---|
| Output not aligned with the brand | High if requirements are vague | Medium | Detailed requirements + sample review |
| Token spending out of control | Media | Medium-high | Monthly budget per agent + automatic alerts |
| Sensitive data exposed to the model | Low with enterprise contract | High | Enterprise contract with the provider, no unverified models |
| Undetected errors in production | Low with review agents | High | QA agent that checks the output before publishing |
| Loss of internal expertise | Long-term average | High | Maintain deliberate friction in the creative process |
Why are people talking about this now and not a year from now?
The Italian AI market grew by 50% in 2025, and Agentic Automation — meaning automation driven by agents that act independently — is already being tested in 8% of major Italian companies. Anyone who waits another year is already falling behind.
The Italian artificial intelligence market reached 1.8 billion euros in 2025, with a 50% growth compared to the previous year, according to the Digital Innovation Observatories of the Polytechnic University of Milan . It is not a niche market: it has become structural.
On the autonomous agents front, data from Intelligent Business Process Automation Observatory of the Politecnico di Milano showing that 8% of large Italian companies have already started experimenting with Agentic Automation in 2025, and 25% plan to do so soon. The gap between large companies and SMEs is huge though: only 8% of small and medium-sized enterprises have started even a single AI project, compared to 71% of large ones, according to data published by AI4Business .
The clearest signal comes from the job market: according to data reported by InnovationCity , in 2025 job postings requiring AI skills grew by 93% in Italy. Companies are no longer looking for people who know how to use a tool: they want those who can build and manage processes with agents.
How to get started concretely: the steps to introduce autonomous agents in a company?
You always start with a process that's already structured, with writeable requirements and an objective quality signal. You don't start with creativity or high-risk processes.
- 1 · Step 1 Map candidate processes List repetitive tasks with clear rules and verifiable outputs. Coding, reports, replying to standard comments, updating product listings.
- 2 · Step 2 Write the requirements as if a new team member were reading them The agent doesn't guess: it executes. The more precise the requirements, the better the output. Test the prompt on real tasks before automating.
- 3 · Step 3 Define the feedback loop Identify the signal that tells the agent if it's doing well: performance KPIs, automated tests, quality checklists.
- 4 · Step 4 Create an active agents registry Keep track of which agents are running, on which tasks, with which model, and at what cost. This is your minimum 'tech council'.
- 5 · Step 5 Include token costs in the operating budget Compare the monthly cost of tokens with the work hours they replace. Update the estimate every quarter.
- 6 · Step 6 Keep deliberate friction on creative tasks Don't delegate everything. Ask the team to think before validating the agent's output. Human judgment is the value that cannot be automated.
What mistakes do companies make when approaching AI agents for the first time?
The most common mistake is starting with the wrong processes: creative or high-risk ones, instead of structured and measurable ones. The second is failing to set up governance before problems arise.
Monks learned the hard way that asking the agent for something 'more daring' produces a lot of noise and little signal. Fifty interesting ideas on paper, but only five truly good ones—and finding those five costs more time than you save. This happens when you use the agent for tasks that require qualitative judgment without having a dedicated expert for selection.
| Error | Consequence | How to avoid it |
|---|---|---|
| Start with creative or high-risk tasks | Unreliable output, loss of trust in the tool | Start with coding, reports, structured tasks |
| Vague or incomplete requirements | Generic output that doesn't save time | Write requirements like a detailed brief |
| No register of active agents | Token spending out of control, no visibility | Create a minimal tracking sheet from day 1 |
| Token budget separated from the operating budget | Invisible costs that grow without comparison | Add tokens to the talent budget or operational budget |
| Delegating everything without deliberate friction | The team loses judgment over time | Keep moments of autonomous reasoning in the process |
| No enterprise contract with the AI provider | Legal risk regarding sensitive customer data | Check MSA and data processing agreement before getting started |
How to tell if the model is working for your business?
Three clear signs: the cost per unit of output drops, senior time shifts from production to curation, and the average quality of the output stays stable or improves.
S4 Capital's EBITDA margin rose to 12.1% in 2025 despite a 10.8% drop in revenue: a sign that operational efficiency made up for the revenue contraction. For an Italian SME, the equivalent KPI is simpler: how much does it cost to produce a unit of output (a report, a product sheet, a post) before and after introducing agents?
| Indicator | How it is measured | Positive signal |
|---|---|---|
| Cost per unit of output | Total € (tokens + senior hours) / number of outputs | Drops by 20%+ compared to baseline |
| Senior time spent on selection vs. production | % of senior hours spent choosing vs. doing | Selection exceeds 50% of total time |
| Average output quality (sample) | Weekly sample review on a 1-5 scale | Stable or growing average |
| Monthly token spend | € tokens / month per agent or process | Stable or declining compared to the plan |
| Undetected errors in production | Number of post-publication reports | Tends to zero with active review agents |
| Team satisfaction | Quarterly survey (1-5 scale) | It doesn't drop: the team perceives the change as positive |
So, is it worth unifying budgets and teams on the Monks model for your company?
It is worth it if you have structured processes with clear requirements, a measurable feedback loop, and a willingness to invest time in governance. Operational savings are real and measurable within the first few months.
A middle ground you are in a transition phase: you have some structured processes but also a lot of creative or high-risk work. Start with the simplest processes, build governance, and gradually expand. Don't wait until everything is ready.
Not yet if your work is almost entirely based on unmeasurable qualitative judgment, or you don't have the resources yet to write decent requirements and manage an agent registry. In this case, it's better to wait and learn first.
How do we see it at SHM?
For us, AI is the means, not the end. The Monks case is instructive precisely because it shows where automation stops: in selection, in judgment, in deciding which direction to take. Fifty ideas in ten minutes are worthless if you don't have someone capable of recognizing the five that matter.
We believe that integrating autonomous agents into a company does not mean reducing people to passive validators. It means giving them back the time to do the work that truly matters: reasoning, choosing, building relationships with clients. The deliberate friction that Monks is reintroducing is not a step backward: it is the recognition that human thought has a value that is not measured in tokens.
In projects where we've supported companies in integrating AI processes, the most critical moment has never been choosing the tool. It has always been defining the requirements: whoever writes the instructions for the agent determines the quality of everything that follows. It's a skill you build, not buy.
If you are considering how to structure this path, you can start with our page dedicated to Strategic AI integration , dive deeper into the topic of business intelligence and data processing or explore options for the development of virtual assistants . You can also follow our hub on AI and process automation to stay updated on how the market is moving.
And if you want to understand how to optimize your company's visibility in AI search engines, the article on OpenAI Astra and computer use offers a useful perspective on where research is heading.
Frequently Asked Questions
It depends on the model used and the complexity of the task. More powerful models consume many more tokens (the processing units) for each operation. Monks' logic is to assign the cheapest model compatible with the task: not all jobs require the best model. Keep track of spending per agent from day one.
Ask yourself two things: can I write down the requirements so that a new team member can understand them without ambiguity? And do I have an objective signal to know if the output is good (a test, a KPI, a checklist)? If the answer to both is yes, the process is a good candidate. If not, work on these two elements first.
Not necessarily, at least in the initial phase. Monks has reduced headcount, but the model it describes is about redistributing work, not a straight-up replacement. People are shifting from production to selection and governance. The risk of losing internal expertise is real in the long run if you don't maintain deliberate friction in the process.
Yes, and some will ask for it. Monks' most advanced clients want an agent registry, to know which models they use, and to be able to 'turn them on and off'. Prepare this transparency before they ask for it: it's a competitive advantage, not an inconvenient obligation.
Before having an agent process sensitive data, check that the AI provider has an enterprise contract (MSA) with clear clauses on data processing and data residency. Unverified models or those without a contract are a real legal risk, as confirmed by concerns from Monks' clients.
On structured processes (coding, reports, product sheets), the first measurable results arrive within 4-8 weeks of going live. Most of the time goes into the setup phase: writing good requirements and building the agent registry. Don't skip this phase to speed things up: it determines the quality of everything else.
Agent autonomy isn't the problem to solve: it's the context that makes human judgment truly valuable.
Related articles
Discover more articles exploring similar topics, selected to offer you a more complete and stimulating perspective. Each piece of content is carefully chosen to enrich your experience.
