Exponential Operator is a 100% human-written newsletter by Sawyer Middeleer, founder of Revi Systems. If you were forwarded this email, join some of the sharpest AI operators by subscribing here:
We AI people, as a rule, are terrible at branding. I've never worked with so many people who have a knack for making something genuinely interesting sound both impenetrably complex and slightly ominous.
Call me quixotic but I've made it my mission to help everyone, not just engineers, understand and make use of advanced AI tools. However inaccessible context graphs, agent harnesses, and loop engineering seem, they're the new foundation upon which companies are being built. (and are really cool, too!)
The latest installment in the cool-yet-ominous AI catalog setting the tech world ablaze is the software factory.
Software factory is actually a bit of a misnomer because the concept as I'll describe it is applicable to any type of organization, not just ones that make software. If you're an operator of any kind, this concept is worth learning about, even though it's still so new that no one agrees on the right way to build one.
I'll do my best to describe what this thing is, and how to approach building one for yourself if you enjoy tinkering like I do. But before that I think it's useful to understand why we're talking about factories all of a sudden. That's exactly the load-bearing piece, as Claude would say.
If you’re not a dev, it can be hard to orient yourself in the world of software. An analogy might help.
Imagine you run an autonomous shoe factory. It's just you and a bunch of robots.
Every morning a pile of raw materials is delivered to your loading bay. A retrieval robot picks it up, and hands it to another robot that sorts it, followed by a third that distributes it around the factory floor to other robots, who load it into their fabrication machines.
These machines crank away, turning rubber into soles and fabric into foot-shaped forms. A little while later, a truck comes by to pick up the finished goods and take them away for sale.
The owner of the factory is somewhere in Mallorca living his best life. But do you, the human in charge, serve any purpose in this arrangement?
When it comes to the production of actual shoes, not really. The factory is fully capable of turning raw materials into shoes all by itself. Inserting yourself into the process would just gum up the works.
Your job instead is to deal with all the much bigger questions. What shoe designs do people want to buy? Is the factory operating to its potential capacity? When does it make sense to add more production lines? Can we coordinate with the hat factory down the street?
In AI bro parlance, your job moved up to a higher level of abstraction. And the software factory is the code version of our shoe factory.
From the aptly named Factory.ai itself:
The software factory is an autonomous system that starts with signals from the outside world, passes them through every stage of the software development lifecycle, and turns them into production software. (post)
It's a neat idea, which we're already seeing take hold in serious product engineering organizations like Linear. User bug report in, software update out, no human in between.
But I'm an operator. So when I first learned about software factories, I immediately thought of how the concept could be applied to everything else a business does, too.
Sales and marketing operations were two of the first areas that came to mind. I think go-to-market activities have a very similar shape to software engineering below the surface.
They both have objective, measurable success criteria. Either the software runs well, or it doesn't. Either you get customers, or you don't. Of course there's a huge space for subjectivity and judgement, but still the most foundational outcomes can be measured by machines.
They both have a transparent and structured ground truth about the work that's being done. The jargony way of saying this is they each have good systems of record. Engineering has a codebase, issue tracking and product usage data. Go to market has customer records, a content library, and correspondence data from emails and call recordings. In both cases, context about the work that's being done is structured and accessible for easy robot consumption.
Finally, they both work on discrete, loop-driven operating cycles that can be completed end to end by AI. Engineering work breaks down into neatly scoped issues - a feature to be built, a bug to be fixed, etc. - where finishing each unit of work follows the same general pattern. Prioritize issues → scope work → implement code change → test & validate → release → measure.
Sales and marketing also work in well-defined loops - campaigns, sequences, lead engagement. Segment a market → identify buyers → create assets → run campaigns → measure results. Over the last 5 or so years these activities have become increasingly digital and data-driven, allowing GTM folks to structure their work similarly to software engineers.
I'm a learn-by-doing kind of person, so flushed with curiosity I naturally started building what you might call a go-to-market factory of my own. Version 1 looks something like this:
I'm not going to spend time in the nuts and bolts of how this thing works. That's for my next post (stay tuned for Part 2!). It's a small factory by any standard, running a simple email prospecting loop. The flow is linear: find companies showing high propensity to need my services, identify the person I should reach out to, send them an email, wait for a response, track results and optimize.
Sales signals in → robots do things → booked meetings out.
I spent 2 weeks building it, which might be surprising to anyone who's set up a similar workflow in hours with Clay. But the difference is I own the whole system top-to-bottom and adding new functionality is just a prompt away.
The entire factory can be run by any agent now - Claude, Hermes, OpenClaw, Codex, you name it.
This pattern is pretty common to run into when you’re building AI ops systems. The upfront work on infrastructure, business logic and guardrails can be time consuming, but once it’s up and running the system pays for itself in a big way.
The most exciting part is that anyone reading this article can build this on a $20 AI subscription (plus some sales outreach software).
Now with Claude the sales rep humming along in the background, I have to find something new to do with my time. Luckily, there's no shortage of demands as a solopreneur.
In Part 2 I’ll share the recipe for exactly how I built this, plus a few issues I ran into and how I tackled them. Subscribe to get it in your inbox!
That’s all for today. Thanks for reading!
江湖再见🫡
-Sawyer