AI works, that's no longer the question.
The question is, how do you turn AI into a business? Over the last year I've noticed something interesting.
AI doesn't become strategic because the models get better, it becomes strategic because new stakeholders enter the conversation.
The engineer asks:
Can we build it?
The CFO asks:
What does it cost?
Treasury asks:
What's our financial exposure?
Operations asks:
Which workflows create value?
The CEO asks:
What's the return?
Those questions create a need for operational visibility.
Visibility creates accountability and accountability creates ownership.
Ownership turns AI from an experiment into business infrastructure, that's the transition I'm exploring.
I'd track policy exceptions, overrides, escalations and repeated interventions, but always with their operational context.
Which workflow? What was the system trying to achieve? What happened, and what was the consequence?
The goal isn't more policies, it's fewer assumptions and better-informed controls.
Ideally, not just an event, but an event understood in context.
The trigger should be a meaningful change against an intended outcome, evaluated against the relevant business context, rules and permissions.
An event tells the workflow something happened, operational memory tells it why that matters, what was intended and whether action is now required.
The customer is no longer necessarily a human, we're moving towards an agent-to-agent service economy, where agents negotiate availability, price, timing and execution on our behalf.
This changes everything around identity, trust, authority and accountability. Finding the electrician is the easy part, governing what the agents are allowed to agree to is the next challenge.
@AditSavani16 Deciding what an agent can access and governing what it can do once it has access.
Permissions define the boundaries; runtime governance ensures those boundaries hold during execution.
@b1rkh0ff Building in public works when it's actually public, not just polished milestones, but difficult decisions, wrong turns, customer feedback and the things you're still figuring out. That's what builds trust over time.
@dominikkoch Probably the best kind of customer research, start looking for ways to help and end up as a fan of the company. Makes the relationship a lot more interesting.
@AIExplainedHQ The same problem appears in agentic workflows, an agent completes the requested task but silently changes something it was supposed to preserve.
Controlling the outcome means making constraints explicit and verifying them throughout execution.
“We'll just make tiny Star Wars characters AI-powered. Nothing major, I promise…”
Yeah, right.
It's probably just the beginning.
How do you feel about your favourite fictional characters becoming AI-powered?
And more importantly… are we ready for an AI-powered Tony Stark?
https://t.co/aISuqzoSBt
@jasonlk Important when you're building in a new category, customers rarely describe the problem in the language you're using to solve it.
Recognise patterns behind what they tell you.
@MediaKing Absolutely and even more important with autonomous AI. They're not just making decisions, nut recognising when the evidence has changed, adapting and understanding the consequences of that change.
@arthuryuzbashev We're getting better at checking what developers ship. Next we need to govern what AI systems do once they're running.
Static security checks and runtime accountability need to become part of the same conversation.
@DanielPriestley Exactly, owning a problem category others still treat as a collection of disconnected features.
In AI, understanding usage is table stakes, understanding the economic consequences of autonomous behaviour is more challenging.
One more number puts this into perspective, Norway's peak electricity consumption is just over 25,000 MW.
The queue plus additional requests for data-centre capacity now total 27,100 MW.
More than the country's peak consumption, requests aren't reserved capacity and reserved capacity isn't connected capacity.
Difference between announced demand, allocated capacity and actual delivery is worth noting.
The AI infrastructure boom has a measurement problem.
We're counting capacity that doesn't necessarily exist in operational terms.
In Norway, Statnett reports:
800 MW of data centres connected.
4,200 MW with capacity reserved.
9,100 MW in the connection queue.
18,000 MW in additional requests.
That's a staggering amount of demand competing for infrastructure that cannot simply be switched on.
That raises a bigger question, are we measuring the AI infrastructure race by what has been announced, what has been financed, what has been connected or what is actually delivering economic value?
These are very different milestones, the next phase of AI infrastructure won't just be about securing more compute, power and capital.
It will be about understanding what those resources are actually producing.
Because capacity is not the same as capability and investment is not the same as returns.
Full story:
https://t.co/R5NydGyHNR
@alexcooldev Spending too much time celebrating revenue milestones and too little time recognising what they actually represent.
The target is making the value repeatable, the economics sustainable and the business scalable.
One founder can now manage a team of AI agents, we know that, but we're starting to create businesses where a significant part of the actual work happens across autonomous systems.
That becomes a new operational challenge, understanding what the agents are doing, how they interact, what they produce and whether the combined output is worth the resources consumed.
@SimonHoiberg Wonder if we'll look back at this period as the transition from being the person who makes things to being the person who manage systems that make things.
Skills don't become worthless, but the advantage moves from execution to judgement.