Hello! 👋
It’s Thursday, 24th September 2026. Hello and welcome back to Bold Efforts!
For most of the last two decades, software companies tried very hard not to become services companies. The logic was obvious. Software scaled, services did not. You could build a piece of software once and sell it to thousands of companies, while a services business had to keep hiring people every time it wanted to grow. SaaS companies got recurring revenue, high gross margins, and wonderful multiples. Consulting firms got timesheets.
AI is starting to make that distinction much less clean. Software has never been easier to build. A reasonably capable person can now take an idea, describe it to an AI, and have something working surprisingly quickly. Features that might once have taken a team several weeks can increasingly be recreated in days, sometimes hours.
That does not mean software has become worthless. Distribution, proprietary data, network effects, deeply embedded workflows, and switching costs still matter enormously.
But simply being able to build software is becoming much less special. And when something becomes abundant, value tends to move somewhere else. In this case, I think it is moving closer to the actual work.
The difficult question for most companies today is no longer whether AI can do something. It is what exactly they should do with it. Which workflows should change? What data should the system have access to? Where should a human remain in the loop? What happens when the model is wrong? How does this connect with the seven systems the company already uses? Who owns it? How do you get employees to trust it? And, perhaps most importantly, does any of this actually improve the economics of the business?
Those questions are much harder than building another application. Which is why an interesting thing is happening across AI right now:
The companies building some of the most advanced technology are moving closer to services.
OpenAI recently created the OpenAI Deployment Company. Its Forward Deployed Engineers are meant to work inside organizations, alongside operators and business leaders, redesigning workflows and deploying AI into actual operations.
Anthropic has taken another route with Ode, an AI-native services company designed to take frontier models and make them useful inside real businesses.
Palantir has operated with this philosophy for much longer, embedding technical teams close to customers and building around the messy particularities of how those organisations actually work.
These companies look different, but the underlying idea is similar. You cannot just hand someone powerful technology and expect transformation to happen. Someone has to go inside the business, understand how it really works, decide what should change, and make the technology fit that reality.
There seem to be two natural directions for AI companies from here.
The first is to move further down the value chain and actually do the activity themselves. Imagine software that helps lawyers perform legal work. If the technology becomes good enough, why stop at selling software to the law firm? Perhaps the more valuable company is eventually the AI-native law firm itself. The same argument can be made for accounting, recruiting, insurance, marketing, research, property management, and dozens of other industries.
Instead of selling the tool, own the outcome.
The second path is to build the AI infrastructure and workflows within the companies. It is what we are beginning to see from companies like OpenAI, Ode, and Palantir. Do not become the law firm. Become the company that goes into the law firm and rebuilds how the law firm operates. That is much closer to a services business than the traditional software model.
For a long time, that would have sounded like a step backwards. The problem with services was always that people were the production function. If you wanted twice as much revenue, you generally needed something approaching twice as many people.
AI changes that equation too. A team of twenty people equipped with powerful models, internal agents, reusable components, and experience from dozens of deployments may eventually be able to do work that once required a hundred. That creates an interesting possibility: the next generation of services companies could combine the intimacy of consulting with some of the economics of software.
They can get close enough to understand the customer deeply, but use AI to make each person dramatically more productive. And because they see the same kinds of problems repeatedly, every deployment can make the next one easier. The distinction between product and services starts to blur.
This also explains why the opportunity is so large. Most companies know AI matters. Far fewer know what to do about it. A CEO can buy access to the best models in the world tomorrow, but that does not tell her which processes should disappear, which ones should be redesigned, what should be automated, or how the company should operate differently three years from now.
The model is increasingly available to everyone but the context is not so.
Perhaps that is where the moat moves: from code to context, from building software to understanding organizations, and from selling tools to delivering outcomes.
For twenty years, the technology industry tried to turn services into software. AI may send some of the value in the opposite direction because when everyone can build software, actually making it work becomes the scarce part. Thank you for reading.
Best,
Kartik
I write Bold Efforts every week to think clearly about where work and life are actually headed. If you want these essays in your inbox, you can subscribe here.