I built an AI-native revenue operating system over the last few days.
Yes, building software has become dramatically easier.
But the more interesting thing happening underneath is a shift in organisational power.
I call the system Pursuit.
Today it can already manage territories, create account queues, research companies, qualify accounts, build detailed commercial cases for whether a salesperson should pursue them, create agentic sales teams, track performance and activity, manage company profiles and provide the beginnings of an integration layer into systems such as HubSpot and Apollo.
There is plenty still missing.
Sales playbooks.
Agent handoffs.
Persistent memory.
Event-driven workflows.
Decision rights.
Governance.
Learning loops.
Strategy simulation.
But building Pursuit made something much larger click for me.
For decades, enterprise software has separated the people who understand how a business works from the people capable of encoding that understanding into systems.
AI is beginning to collapse that separation.
And that has consequences far beyond productivity.
It changes who controls the machine.
The hidden hierarchy inside a revenue organisation
I have worked around revenue organisations from several different angles.
I’ve worked in solution architecture and pre-sales.
I’ve worked alongside sales organisations.
I’ve built channel partner businesses.
I’ve managed programs, commercial outcomes, operating models and complex organisations.
One thing becomes obvious after spending enough time inside these environments:
A sales organisation is not really a group of salespeople.
It is a system.
There are territories.
Accounts.
Segments.
Qualification rules.
Partner relationships.
Incentives.
Targets.
Approval structures.
Escalations.
Pricing constraints.
Sales stages.
Political relationships.
Customer intelligence.
Management judgement.
There is a machine underneath the people.
The problem is that very few companies actually control that machine end to end.
Parts of it live in Salesforce.
Parts live in spreadsheets.
Parts live in PowerPoint.
Parts live in the head of the VP of Sales.
Parts live with RevOps.
Parts live inside marketing automation platforms.
Parts live inside the experience of the salesperson who has been covering the territory for six years.
Parts live nowhere.
And then we wonder why organisations lose knowledge every time someone resigns.
The modern revenue stack solved one problem extremely well.
It recorded things.
It did not necessarily make the organisation intelligent.
CRM won the first war
The CRM became one of the most important pieces of software inside the modern company because it became the system of record.
Who is the customer?
What is the opportunity?
How much is the deal worth?
What stage is it in?
Who owns the account?
When did someone last contact them?
That database became enormously valuable.
Once a system becomes the authoritative record of commercial activity, switching away from it becomes extremely painful.
That created tremendous power for companies such as Salesforce.
But recording a decision and making a decision are two very different things.
The next layer of enterprise software may not compete with the CRM on storing customer records.
It may sit above it.
It will decide:
Which account should we pursue?
Why?
What changed?
Which salesperson or agent should handle it?
What should happen next?
Should the account receive more investment?
Which stakeholder should we contact?
Which playbook should we run?
When should a human intervene?
What did we learn from the previous hundred decisions?
That is a fundamentally more powerful position in the stack.
The CRM tells you what happened.
The operating system decides what happens next.
The AI opportunity is not automation
A lot of AI discussion still revolves around automation.
Can AI write an email?
Can AI research a company?
Can AI update Salesforce?
Can AI create a proposal?
Useful, certainly.
But those are tasks.
The larger opportunity is orchestration.
Think about the entire commercial system:
Territory → Account → Research → Qualification → Decision → Action → Pipeline → Learning
Most companies currently operate each part separately.
An analyst researches.
A salesperson qualifies.
A manager prioritises.
RevOps builds a workflow.
Marketing provides intent data.
Someone updates the CRM.
Management reviews the pipeline.
Then everyone meets on Monday morning to reconstruct what is happening.
That is not an operating system.
That is human middleware.
AI potentially removes some of that middleware.
Not necessarily the humans.
The translation.
This is where the power shift happens
Historically, if I understood exactly how I wanted a revenue organisation to operate, understanding was not enough.
I needed software architects.
Product managers.
Developers.
Data engineers.
Integration teams.
Security.
RevOps.
Budget.
Months of translation between people.
Every translation layer diluted the original operating model.
Eventually you received software that approximated the process.
That constraint created a particular kind of organisational hierarchy.
The people capable of building systems had enormous leverage over the people who merely understood the business.
AI changes the economics of that relationship.
Someone who understands the business deeply can increasingly move from:
I know how this should work
to
I can encode how this should work.
That is an enormous shift.
The scarce skill stops being pure software production.
The scarce skill becomes knowing what system should exist in the first place.
Systems thinking becomes more valuable, not less
There is a slightly strange assumption floating around that AI coding tools reduce the value of expertise.
I think the opposite is happening.
Coding becomes cheaper.
Judgement becomes more expensive.
When almost anybody can generate an interface, the question becomes whether the person behind that interface understands the system.
What are the decision rights?
What should be automated?
What should remain human?
What is a good account?
What does “qualified” actually mean?
When should a salesperson stop pursuing an opportunity?
Which events should trigger action?
What information should persist?
What should the AI be allowed to do without approval?
What happens when incentives conflict?
How should the system learn from failure?
These aren’t coding questions.
They’re organisational design questions.
And that is why I don’t find “AI lets everyone build apps” particularly interesting.
The important question is:
What happens when domain experts can finally encode their own models of reality?
Revenue teams have spent years adapting to their software
This is the part I think is underappreciated.
Software is never neutral.
Once a company adopts a system, people start reorganising themselves around it.
Sales stages become the stages available in the CRM.
Reporting structures follow what dashboards can measure.
Management starts caring about metrics because they are easy to extract.
Processes change because the workflow software prefers certain sequences.
Eventually the organisation forgets whether the software reflects the business or the business reflects the software.
The tool gains power.
This is one reason enterprise SaaS companies became so valuable.
They didn’t merely sell software.
They embedded themselves into the operating model of the customer.
AI-native software has the potential to reverse some of that relationship.
Instead of asking:
“How does Salesforce think our sales process should work?”
a company can increasingly ask:
“How do we want our sales organisation to work?”
And then encode that.
That gives some power back to the operator.
Pursuit is my experiment in this idea
That is what I’m testing with Pursuit.
I’m not particularly interested in building another CRM.
There are enough CRMs.
I am interested in what happens above the CRM.
The current version already provides a basic command centre.
It has territories.
Account queues.
Sales agents.
What comes next is more important.
I want playbooks that don’t sit inside PDFs.
I want executable playbooks.
I want the system to understand:
Here is the objective.
Here are the constraints.
Here are the accounts.
Here is how we decide.
Here is what AI can do.
Here is what requires human approval.
Here is what should happen when the environment changes.
Then agents can operate within that structure.
One agent discovers an account.
Another researches it.
Another qualifies it.
Another builds the commercial thesis.
Another prepares the action.
A human intervenes where judgement matters.
The system remembers what happened.
Then the system learns.
That begins to resemble an operating model rather than a collection of AI features.
Control is the point
There is a tendency to describe agentic AI as giving up control.
Give the machine an objective and hope it does the work.
I think serious enterprise AI will move in the opposite direction.
The winners will build systems with much more explicit control.
Goals.
Constraints.
Authority levels.
Approval gates.
Decision logs.
Auditability.
Escalation rules.
Economic limits.
Humans will not disappear from the loop.
The loop itself will become much more visible.
That may actually give management more control over complex organisations than they have today.
Because right now, a huge amount of decision-making is invisible.
It happens in inboxes.
Slack messages.
Phone calls.
Individual judgement.
Spreadsheets.
Memory.
A properly designed agentic operating system can make those decisions explicit.
Who decided?
Based on what?
Under which rule?
Using which evidence?
Was the decision overridden?
What happened afterward?
That is not merely automation.
That is governance.
And governance is power.
The winners and losers are starting to become visible
If this direction is correct, there are some uncomfortable implications.
The winners will not necessarily be the companies with the largest collection of AI features.
They may be the companies that control the orchestration layer.
The layer sitting between company strategy and execution.
Traditional CRMs probably remain important because systems of record are extremely sticky.
Data vendors remain important because agents need high-quality external intelligence.
But entire categories of workflow SaaS could come under pressure.
If companies can increasingly encode workflows themselves, paying another SaaS vendor for every individual business process becomes harder to justify.
There is another potential loser.
Organisations built around translation.
People whose primary value is transporting information between systems, departments or hierarchy levels may find that part of their role increasingly automated.
Meanwhile, domain experts who understand the complete system may become significantly more powerful.
The salesperson who understands only selling may not gain much.
The developer who understands only coding may not gain as much as expected.
But the operator who understands customers, economics, process, incentives, technology and organisational behaviour suddenly has extraordinary leverage.
That person can begin to build the machine.
The full-stack operator
We talk about full-stack developers.
I think AI creates another archetype.
The full-stack operator.
Someone capable of understanding strategy at the top, systems in the middle and execution at the bottom.
Someone who can move between commercial logic, organisational design, technology and implementation.
Historically, organisations separated those skills because no individual could realistically execute across the entire stack.
AI changes that constraint.
That doesn’t mean everybody becomes a one-person company.
It means a single person can now operate across boundaries that previously required whole departments.
That changes the leverage of individuals.
It changes the economics of startups.
And eventually it changes the structure of large companies.
Software is becoming less important. Systems are becoming more important.
That sounds contradictory in the middle of an AI software boom.
But I think it’s exactly what is happening.
When software becomes easier to produce, software itself becomes less scarce.
The system behind the software becomes the valuable thing.
Your understanding of the customer.
Your model of the market.
Your decision rules.
Your institutional knowledge.
Your proprietary data.
Your workflow.
Your organisational judgement.
Your operating model.
Those become the moat.
Code increasingly becomes the expression of those things.
That is the lesson I’m taking from building Pursuit.
Yes, AI let me build something in days that would previously have taken considerably longer.
Interesting.
But that’s not the story.
The story is that the distance between understanding the machine and building the machine is collapsing.
And whenever a layer of translation disappears, power moves.
The question is where it moves next.
My bet is toward the people who understand the whole system.
And know what to do with it.