You're in your leadership meeting having just sketched out a new strategy on the whiteboard. Everyone nods. But then comes the question that always comes: who owns making sure this happens? And who tracks what actually happens versus what we planned?
Most answers are: we hope someone remembers. Or we do weekly project-manager check-ins. Or we have a controller who does quarterly reviews that arrive three months too late anyway.
But what if you had a department that actually worked on it?
Not another person. Not more meetings. Just a structured group of agents, each knowing their job, working in parallel, reporting back without needing permission every time.
That's how AI agents work in practice.
The 6 departments
When you fire up a strategy board in 360° Sprint, the work gets distributed across 6 agent departments — each with a clear mandate.
Strategic Intelligence gathers the context. What's happening in the market? What did we miss? This department reads your sources, scans the news, and compiles a situation report before the rest of strategy even begins.
Analysis interprets what Intelligence found. What does it mean for us? Where's the risk? This department takes raw data and builds a narrative — the connective thread that makes it all clickable for leadership.
Development designs the options. Given what we now know, which 3-5 courses of action are worth considering? This is the creative department — the one that asks "what if we did it this way?"
Direction prioritizes. Of all the options, which one do we follow? Which risks can we live with? This is where the human needs to be — the human who knows the organization from the inside.
Execution plans the work. Given we're going this direction, what needs to happen by whom and when? How many resources? What can go wrong?
Facilitation keeps score. The Monday meeting got rescheduled? Noted and timeline adjusted. A critical metric didn't get measured? Flagged.
The key isn't that each department is perfect. The key is that each knows its job, and they work without stepping on each other.
Topological sorting: How it all fits together
When you click "Run strategy" it looks clean on the screen — some boxes turn green, some get data filled in, some wait for input from other work.
Under the hood something more precise happens. Each node (each decision, each analysis, each plan) depends on some other nodes. Strategic direction can't be prioritized until Analysis has told leadership what the data means. Execution plans can't be built until Direction has chosen which path.
The system that orders all this is called topological sorting. Simple version: first the things that need no other input, then what can be based on that, then the next layer. Like building a house — foundation first, then walls, then roof. Not backwards.
This means even if your board has 40 nodes across all 6 departments, the system automatically knows what can work in parallel and what's blocked. And it respects those constraints without asking you every time.
Budgets: Why AI doesn't run wild
Here's where a lot of companies get nervous. If I have 6 agents working 24/7, doesn't that get expensive?
Yes. If you don't set boundaries.
So boundaries are the first thing you set up. A budget: 100 dollars per strategy run. Per week. Per agent role. Whatever you choose.
When an agent uses tokens (every time it reads something or writes something) that comes out of the budget. When the budget runs out, the agent stops. It finishes its work and waits for next period.
This doesn't mean you lose work. It means the agent works within constraints — the same way any team member does.
Concretely: if your Analysis department is burning through tokens reading too many external sources, you set a cap: "maximum 30 sources" or "maximum 3,000 tokens per analysis." The agent adapts.
The key point is budgets aren't prohibitions. They're guardrails. Like when a marketing manager says "we don't spend more than 50,000 kr on ads this week" — the details are her call, but she knows where the line is.
A concrete example: From idea to execution plan
Say your leadership team needs to decide: should we be on TikTok?
Monday morning, day 1. You set up the node on your board: "Evaluate TikTok as a distribution channel."
The system moves immediately.
Intelligence agent searches: Who's selling our category on TikTok? What does TikTok's own data say about user age distribution? What did competitors try? The agent starts now — while you're still in the meeting.
Analysis agent waits for Intelligence to finish (see topological sorting). Once it does, the agent reads what Intelligence found. Writes a short analysis: "TikTok users are mostly under 30. We don't reach them. Competitors who tried it stopped after 3 months."
Development agent reads the analysis. Designs three scenarios: 1. We start a test channel, 3 months, budget 5,000 kr/month 2. We hire an agency to run it, same budget 3. We don't do it now, evaluate again in 2027
Direction agent — and here it stops. This needs human judgment. The agent has prepped all the information. But the call is yours.
Normally you'd say: "Option 1. We test for 3 months."
Execution agent springs into action. Who does this? What resources? When do we start? What are KPIs? It builds a timeline, assigns tasks to people you've named.
Facilitation agent logs all this and sends a weekly digest every Monday: "TikTok test launches next week, first post Wednesday, first data in by April 30, review meeting April 20."
The entire cycle took less time than an actual meeting would have.
And if a meeting gets rescheduled? The facilitation agent handles it automatically.
Why this beats "ask ChatGPT"
Right now, when a leader has a strategic question, they sit down and type a prompt into ChatGPT. ChatGPT gives an answer. They read it, judge it, remember it (or forget it).
The problems:
- ChatGPT doesn't know what you learned last week — it's stateless
- ChatGPT has the same context every time — it doesn't learn
- It requires you to remember to ask — if you forget, no answer
- ChatGPT can't weight your strategic priorities — every question ranks equally
Agents solve all four.
An agent remembers what was decided last week (Knowledge Graph). An agent learns what worked (if last quarter's priority delivered X%, note it). An agent works continuously — you don't trigger it. And an agent knows what ranks highest because you configured it.
Another way to put it: agents are ChatGPT actually working for you, not just answering questions.
The only executive team that never sleeps
If you use agents right, you have what amounts to a strategic department that works without asking permission. It reads what's happening in the market, judges what it means, writes down what you should consider, prioritizes what's critical, plans the work, and keeps score on whether it gets done.
That's not a fit for regular meetings. That's how strategy should work — as continuous work, not an annual exercise.
It doesn't mean you skip meetings. It means when you meet, you have actual data to discuss instead of spending half the time building context from scratch.
If that sounds like something your company could use, look at 360° Sprint. And if it sounds abstract, run a strategy — then you'll see exactly how it works.
A department that works without asking permission. That's what agents can be.