# AI Won't Make Everyone A Tool Builder

**Podcast:** Another Podcast
**Published:** 2026-07-24

## Transcript

Hi, I'm Tony Karen Brown.
And I'm Benedict Evans.
You wrote a piece about new tools, and that doesn't mean that automatically people will become tool builders.
Well, it is.
So there's this sort of weird narrative that says you've got Claude Cowork.
Yeah.
And you won't need your software and it will change.
You won't need software anymore and it will change how you work and it will transform.
how you do what you do, and everyone's going to have this.
And I kind of looked at this a little bit and scratched my head and thought, so is this, you mean the way Microsoft Access destroyed databases?
Which is sort of a facetious but kind of relevant comment.
And I thought there's sort of three building blocks to think about here, which is that, first of all, like most people don't see the problem that you're going to automate.
Most people could not build that tool, even if they did see the problem.
do not think in the right ways that you need to think to build that tool.
And that's got nothing to do with writing software, nothing to do with writing code.
And most people are not in a position that they could get that deployed and used, even if they did see the problem and even if they were able to think about the right way of building this thing.
And so if one sort of unpicks each of those things...
If you actually go and kind of remember how most software works, like how many of the tools that you use today were something where when you first saw it, you didn't get it.
And you thought it doesn't make any sense.
I would never do that.
And that applies.
I don't need that.
I don't have that problem.
I don't even have that problem.
Never mind that I would use this.
And so most people are not sitting thinking, how could I automate this task?
How could software change this?
How could this be improved?
And of course, if you're a software developer, that's all you think about.
And so this is like an alien mentality.
But most people are busy doing their actual job and not thinking about how they would automate it.
I think the second and kind of perhaps more useful point is that there's a huge difference even between seeing the problem and being the right person to work out like.
what should the processes and the workflows and the networks and the functions be that would make this great?
And particularly a whole bunch of categories where like 10 people have tried to do that and failed already.
So very often it's not obvious how this thing should work or it looks obvious.
But the mark of great software is you see something and you think, oh, wow, that's a great idea.
And you can't.
And so then if you're a great accountant or a great video editor.
or a great graphic designer, or a great salesperson, or a great lawyer.
Those are different skills to being the person who's going to work out the right way to make a really great piece of legal discovery software, or a really great piece of sales enablement, a really great sales enablement tool.
Like, you'll have that problem every day, but you won't see it.
You won't realise that you could fix it by turning it into this problem and then building a thing that did that.
There's different skills.
And those are different people.
And then the third building block is most of these tasks are not things being done by one person by themselves.
Yeah.
Some of them are, but most of them you're in a team of 50 or 500 people and the data is regulated and the data touches three or four different systems of record that need to be tightly controlled and have security and compliance and audit.
And so you can't just get your whole company to go off and use this thing.
Or at any rate, you need to get some approval from somewhere to get people to use this thing.
And it's across different teams who have different processes and different ways of working.
Across different teams, maybe different processes, different stakeholders, different budgets, different companies.
And so the example I sort of thought about here was something we looked at when I was at Andreessen Horowitz called Frame.io.
which is like a professional video collaboration platform.
So there's a piece of video being made.
There's like half a dozen people and maybe three different companies who are actually working on it.
But then there's 20 or 30 or 50 people across another three different client side companies who have to see it, have an opinion on it, sign off on it, check it.
And so before frame.io, that would have been a bunch of private Vimeos or Dropbox links or maybe some FedEx hard disks.
And then it would have been an email thread.
And if you were lucky, it would be a Gmail, a shared, a shared.
a shared Google sheet with time codes in one column and then comments in rows.
And frame.io says, no, we're going to do version tracking.
And you can go to that frame and draw a circle over that frame.
And you have permissioning and tracking and all the stuff you can imagine you would want there.
Now, if you are the colorist or you are the person doing review at the ad agency, but not the client or the marketing CMO at the client.
You're not the right person to see all of those problems.
You're also not the right person to get everybody to use this.
And so there's just this kind of huge gap between we give everyone a tool that can make stuff and therefore everyone is going to make new tools.
this idea of like getting the other 49 people to use it and adopt it, is that still a bottleneck where potentially in this day and age there is no it to adopt?
Just every person can sort of create their own assistant to do the connecting work and the building blocks.
Everyone can create their own agents to help.
Like, do you know what I mean?
Like, have we shifted in that everyone can build their own agent so there's no sort of central it to adopt because we've now got agents talking to each other?
So I think the Well, I think the answer to that, and this kind of comes back to the counter argument to all this, is how many, like, 50-person company departments run on a 10 meg Excel file?
And the answer is quite a lot.
A while ago, as a quote I made into a slide, somebody told me on Excel that half, someone told me on Twitter that half of their jobs were telling people to use Excel to use a database and the other half were telling people to use a database to use Excel.
And so there is, yes, clearly there is shadow IT.
Like, you know, there's a whole bunch of software.
You know, every big company has five different MailChimp accounts.
But that's slightly different.
That's not somebody in marketing said, hey, I'm going to build my own marketing emailing software.
They went out and bought the software and then their team is using it.
That doesn't change how that doesn't replace the big iron software that the company is already using.
So, and, you know, the typical big company today has four to five hundred SaaS apps and many of those are bought bottom up.
But they're not built bottom up.
And the reason they're not built bottom up isn't that it's hard to write the code necessarily.
it's that there are all these other strictures in how you would think about what it should be doing and how it should work now go back to that 10 meg excel file that a department is running on no one actually knows who built it no one really knows what that sheet is doing right there's a whole bit of it that's blocked off in red it's like don't touch this we don't know why it's here but it could break the whole thing Yeah.
And there's always a point where the company says, no, it's time to move this to software.
I mean, this is kind of my point about institutionalized software versus improvised software.
There's a point where you do it in Excel and there's a point where you buy a thing because you want to know that somebody's thought about how it should work and somebody is responsible for it and will fix it and there will be bug fixes and security.
Can you argue that people who were built and who are using the Excel sheet, though, are kind of tall, tall builders themselves?
Or do you think that's pushing the narrative a bit?
The question is how many people do that?
How many people were actually building the big Excel, were building the Excel sheet to do all of that?
And, you know, there's clearly a kind of big kind of fuzzy area.
Is it a spreadsheet database?
Sometimes, yes.
What exactly is the dividing line between a...
500k, Excel file, and SAP.
Well, there are a couple of kind of breakpoints along that transition, but there isn't kind of one point where you say this is no longer a spreadsheet, this is now a database.
I mean, you know, I'm sure there's somebody screaming at this saying, no, no, it's about joins and tables and things, but like there's a sort of fuzzy continuum between a Word document and SAP.
There's a Word document that contains your holiday policies.
At a certain point, there's a...
a system in Workday that tells you what your holiday allowances are.
And yes, AI kind of moves those back and forth.
But the thing I kind of wanted to get to is like starting at the beginning, most people don't see that you could do that in software.
Don't see, oh, we could change how you do this.
And because they're busy doing something else, they're busy being really good at sales.
They're busy being really good.
I mean, this is also kind of the conversation about Notion.
which is how many people think in terms of, oh, I'm going to build a notion to run my life.
And of course, if you're in tech, the answer is everybody, everyone does that.
What do you mean you don't have a notion board running your life?
Well, a Kanban board, like what do you mean you're not running in this?
And the reality is there's a certain kind of person that thinks like that, that meshes with a certain kind of job.
You bring up Slack and Notion a lot as sort of these outliers.
I'm curious.
What do you think the two have in common that let them be the agglomerate?
Most people aren't tool builders.
Well, so obviously one person can build the notion that other people use.
And it's a lightweight collaborative database system that doesn't otherwise have kind of good parallels.
Slack had a network effect.
in that one team would use it and then someone was working with that team so they would use Slack and then their team would use it and then they would kind of spread organically.
There were very, very few pieces of enterprise software where that's actually happened and where that's worked.
It's like there's Slack and there's like one other I forget.
And what this is, you know, it's kind of the point I made at the beginning of the essay was there was this whole story kind of 10, 15 years ago in the Valley that like...
enterprise software will become grassroots.
It'll be bottom up, bottom up enterprise, bottom up sales.
You won't have to go through the sales process.
You won't go to CIO.
You won't have to wait to go through an 18 month sales cycle.
The users will see the thing and buy it.
And it turned out that that gets you 5% of the market.
And for everyone else, no, actually you need to go and evangelize.
You need to explain to people why this is a thing and why they need it and why they need to budget for it and why they need to buy it.
And you need to explain to people why this problem exists.
And then you need to show them why your thing solves it.
And generally, five other people have tried to solve it before and failed, which is another problem with it, where people will just build their own tool.
They'll build a tool that they don't use it so it doesn't work, that doesn't solve the problem in the right way.
And so this is, I suppose, a sort of a high level.
point in here somewhere, which is, you know, it's a thing I kind of keep coming back to is what are all the ways that AI is, this AI thing is kind of completely new and different from anything that's happened before.
And what are all the ways?
And it's kind of, it's a new flavor on what happened 10 years ago and 20 years ago and 30 years ago.
You know, I do keep mentioning Notion.
Notion is a no code story.
No code was a story from 10 years ago.
No code in general.
worked up to a point but it still doesn't sell itself just because it's easy to make and easy to use exactly and go back in our previous generation i mean i mentioned in part of the passing access which i don't think access even exists anymore but like nobody will need databases anymore because you can just make your own with access and i made a access database for my mother's publishing business it was great but that didn't replace every other piece of software or neither did excel excel didn't replace accounting software Wrong timeline there.
But we have both Quicken and Excel.
Most people don't do the taxes in Excel.
Most people use QuickBooks because it's more complicated than that.
And so I suppose there's a point here, which is this will be more software and people will use this to make tools the way they did with no code and the way they did with Excel and the way they did with all those previous things.
But it doesn't actually fundamentally change the dynamic.
It just shifts those thresholds another level.
And it's interesting because if you're saying, look, AI doesn't eliminate the need for someone to see the problem and you still then need to sell the solution internally and get everyone to adopt it.
Are we shifting the value then back to product and go to market strategy, which is what we always talk about?
Like the differentiator isn't going to be how good the product is and what it does, but it's going to be adoption and like selling it as this sexy new tool that you all need to have.
Like, is it a fight of better engineers or better marketing?
why people love this word taste all of a sudden.
Yes, the tastemakers of the world, Benedict.
But the decision, well, no, it's like the decision of knowing, if writing the code, writing the code, sorry, writing the code isn't the hard part.
Writing the hard part is knowing what the code should be doing.
It's seeing the problem and knowing the right way to do it.
It's the opinion.
And there's many different ways you can look at this, but in principle, what an LLM does is it tells you how most people will probably do that.
And so is that what you want?
Or do you need to think of a new way of doing this?
Or have you thought of a new way of doing this and you want that?
Well, the LLM can do it if you can explain it, if you know how to explain it, if you are the kind of person that knows what software is and how it works and what the things you might ask for might look like.
But if you don't know what those kind of modalities are called and how they work and what people have done before, then you won't sort of sit and think, oh, well, you know, what we could do is we could take that thing that used to work for this in this industry and we could change it like that and we could make it work here.
If you're a really good lawyer, you haven't seen all of that stuff.
You don't.
That's not what you spend your time thinking about.
Do you think that becomes then a de facto new skill set that you can be an incredible lawyer?
But if you don't understand how to think through the next generation of tools, then maybe.
I was going to pull it.
I was going to pull that in another way, which is that.
The lawyer has all these new AI tools.
And so what matters is your opinion as a lawyer, as opposed to doing the law the way everyone does it.
As long as not the process, but how you get to it.
Yes.
Your value as a lawyer is doing something that isn't exactly how anybody would do it.
Your value as a developer...
as writing the code as you want the code written the way anyone would do it but the product the point of the product is thinking of a product that isn't how everybody would do it and as a lawyer your skill is in having opinions about law it's not in having taste in opinions about software It's interesting, I'm seeing that even as my stuff in Formula One used to be very factual, and now the reality is good facts and the most adequate, or the most recent facts are easier and easier to get, that people are actually asking for me for my opinion on something, and how I would solve a solution, which are things that I would normally never touch.
But it must be the same for you as well, like it's easier and easier to put together a newsletter, but actually to have a stark opinion that...
drives a conversation is more valuable.
Yeah, so a billion people are using AI to make newsletters.
Is that true?
What is your voice?
Is that a natural stat?
Billion.
No.
No, I'm just metaphorically speaking.
Don't be so strange.
You can't just throw a stat and a number at me and then click that.
Well, metaphorically speaking, lots of people, there are an enormous number of robo-newsletters.
But then the narrow problem is they're not very good, but the deeper problem is they all say the same thing.
They all say what anybody would probably say, and that has no value.
Well, it has value for one of them, but not for millions of them.
And this is, you know, back to my point, if you have great taste and opinion, if we're going to use those terms about how to do really good sales, that's a different thing to having good taste and opinion about what enterprise sales software should look like and what new problem you could solve with it.
As I said, most people aren't tool builders.
The people who are good at using the tool are not the same people as those who are really good at creating the tool.
Now, sometimes it's the other way around.
Like to make good software to do that thing, you have to know a lot about that thing.
But those aren't the same point.
You have to know a lot about sales to make good sales software.
But being good at sales does not mean you're going to make good sales software.
Okay, there we are.
I've explained the world.
It's funny, though, because it is this sort of recurrent thing every 10 or 15 years.
Like, no-code people will make their own stuff.
And it's something you see very clearly now around co-pilot and co-work and what's the word?
ChatGPT work.
And chatGPT OpenAI just suddenly replaced everybody's chat with this weird, bungled up, completely chaotic and confused thing called work.
And it's like, well, what is this and why does it work?
Because that's what everyone will be using.
Everyone will be writing their own code.
And it's interesting.
I did have, I think, this mentality or this thought that the hard part was actually building the code, but it's actually not the code.
It's knowing, as you're saying, it's understanding and knowing the tool should exist in the first place and then getting everyone else to use it, which I agree with the last bit.
That has always been my struggle.
And so now I'm thinking, well, in a world of AI where you can build your own agents, are we still going to have to worry about adoption and convincing other people to use it?
Yeah, because people won't see.
the thing that they want the agent to do and then they won't be able to work out how they want the agent to do it.
They won't be able to articulate that.
Most people can't do a flowchart of how they do their job every day.
It's still a thing that people have a hard time doing.
Yes, it's funny.
People say prompt engineering is gone.
Bullshit.
The underlying challenge is how do I work out how I would tell the model how I want to do this?
And it's not like you have to say clever things to get the model to work in a different way.
It's like, how would I tell a 19-year-old?
or 25 year old, how I want to do this?
And how would I know that I want to do that?
And how would I know that problem exists?
And do I have the kind of tasks that are obvious things that you could automate?
You know, some people have a bunch of really repetitive tasks that they do every day, and it's kind of tedious and pain and time consuming.
And they say, oh, I could get AI to do this.
A lot of people don't have like a specific thing they do every day that's tedious and repetitive.
Or they may do, but they don't realise that they have it.
They may not see that that thing is there.
Yeah, because they've always done it that way and most people don't.
They've always done it that way and they haven't realised that they're doing this tedious automatic thing because they're kind of not doing it.
Or they've never done it because it couldn't be done or because it would be tedious.
And now, and then you give them a piece of software that says, wait, but I can go and do that for you.
It's seeing the problem and then realising what the tool would be that would solve that.
And then getting everybody to do it is quite different from giving everyone a piece of a thing that can make tools.
Those are different people.
Yeah, because the tools might be there to make the new tools, but the reality is the people haven't changed is what we're getting to.
The software has gotten better and better and the no code tools are there, but the reality is we haven't changed.
We're still the same humans that we want.
I like that.
Okay.
Okay.
Like that.
There we are.
Have a nice day.
We should pad that.
We need to pad.
I was going to say we need to pad this out for another two hours.
It's going to be a proper way I broadcast.
Short and succinct.
This is good.
Short and succinct.
20 minutes.
Three questions.
Three answers.
I like that.
Talk to you soon.
Great.
Speak to you later.
Bye.