Lately, I’ve noticed something important in client conversations.
Almost all of them already know AI can code well now.
That is no longer the interesting part.
They don’t really care whether Claude, Cursor, or any other tool can build an app, write functions, or scaffold a dashboard. That has already become normalised.
Their concern is still the same as before:
- Can technology help my business?
- Can it make a workflow faster?
- Can it reduce back-and-forth?
- Can it improve service quality?
- Can it save my team time without adding more complexity?
That is why I think a lot of AI discussion is still pointed at the wrong thing.
The market is obsessed with whether AI can code. Buyers are still focused on whether software can help them operate better.
Those are not the same question.
AI coding is impressive, but it is not the product
AI coding is real. It matters. It changes how software gets built.
But for most clients, that is not the value proposition.
No one buys software because they are impressed that it was built quickly. They buy software because it helps them do a job better.
This is why the a16z podcast conversation with Atlassian’s Mike Cannon-Brookes clicked for me. The most useful framing in that discussion was this:
The whole history of software was turning filing cabinets into databases. Now the filing cabinet can do work.
That is the real shift.
The old software model was about recording information. The new software model is about helping move work forward.
What clients actually want
When I speak with clients, they are usually not asking:
- Can AI generate code?
- Can it build an app from a prompt?
- Can it replace developers entirely?
What they really want to know is:
- Can this help my support team resolve issues faster?
- Can this reduce manual ops work?
- Can this speed up approvals and internal coordination?
- Can this help my team produce more without hiring too fast?
- Can this fit into the way we already work?
That is a workflow question, not a model benchmark question.
And that is why I think many AI products still feel more exciting on Twitter than in actual businesses.
They demo well. But they are not yet clearly attached to a painful enough business problem.
From systems of record to systems of process
For years, software mostly digitised information.
- HR files became Workday
- support tickets became Zendesk
- accounting records became QuickBooks
- CRM notes became Salesforce
That was a huge step forward, but it still left humans doing most of the actual work around the data.
People still had to:
- read the context
- decide what mattered
- route the task
- follow up with another team
- approve something
- summarise the issue
- move the process along
This is where AI starts to matter.
Not because it can write code, but because it can help software become a system of process, not just a system of record.
That is much closer to how clients think.
They do not wake up wanting a smarter database. They want fewer bottlenecks. They want less repetitive work. They want the business to move faster.
Why I don’t buy the “SaaS is dead” take
I do think some SaaS products will get squeezed.
Especially the ones where the value is thin, the logic is shallow, and the pricing is mostly attached to human seats doing repetitive tasks.
But I don’t think serious business software disappears just because AI can generate interfaces and code.
The reason is simple:
Most business software is not just UI plus a database. It is years of accumulated workflow logic.
It contains the messy parts:
- approval rules
- compliance steps
- regional exceptions
- handoff logic
- finance controls
- support escalation paths
- permissioning
- reporting requirements
That is the stuff clients actually depend on.
You can vibe code a demo in a weekend. You usually cannot replace years of embedded business logic and edge cases that easily.
So my view is not that software gets replaced by prompts. It is that software gets re-rated based on how much real workflow value it owns.
The hard part is not intelligence. It is trust.
Another thing clients care about a lot more than the AI crowd sometimes admits: trust.
Even if the AI is capable, the business still has practical questions:
- When should it act automatically?
- When should it ask for approval?
- How much should it show?
- How do we know what it changed?
- How do we stop it from becoming another layer of confusion?
That is not a raw model problem. That is a product and workflow design problem.
This is where I think a lot of the real value will be created.
Not by the teams that shout the loudest about autonomous agents. But by the teams that quietly design software people can actually trust inside a real business process.
The best AI features may look boring
I also think the most valuable AI features, at least in the near term, may look quite boring.
For example:
- summarising a long support ticket properly
- drafting a follow-up based on internal context
- routing a request to the right team
- highlighting risk before approval
- preparing a clean brief before a meeting
These are not flashy demo moments. But they are exactly the kind of things clients care about, because they save time in places where teams already feel pain.
That is why I think many successful AI products will win first as workflow assists, not magical all-in-one agents.
My current takeaway
My biggest takeaway right now is this:
AI coding has already become table stakes in people’s minds. It is impressive, but it is not the main business question anymore.
The real question is still the old one:
How does this technology help the business?
That is the filter I increasingly use now.
Not:
- how magical is the demo?
- how good is the code generation?
- how many agents are involved?
But:
- does it solve a real workflow problem?
- does it fit how a team actually works?
- does it reduce effort, delay, or confusion?
- does the buyer understand the value quickly?
If the answer is yes, then AI makes the software more valuable. If the answer is no, then it is probably still just a good demo.
That is why I don’t think AI kills SaaS.
I think it forces SaaS to grow up.