12 Factor Companies

In the spirit of 12-factor apps, and 12-factor agents - I propose 12-factor companies. This document is a living document - I’ll continue to update it.

Organizations shape technology and technology shapes organizations. Double-entry bookkeeping, the printing press, steam power, fax, the internet and software of the past 30 years have each changed what it means to build and run great organizations. AI will do the same.

Though we are still very early in this arc - it will come swiftly with a lot of creative destruction. It is worth trying to predict how organizations of the future will operate.

The firm of the past was designed around the bottleneck of human execution. The firm of the future is designed around the bottleneck of human taste and judgement. If execution is largely commoditized - then context, knowledge, secrets and taste will accrue value. The old firm bundled labor and coordination. The new firm bundles context and authority.

Data vs compute

• (1) Own your context
• (2) Rent your intelligence

Build vs buy

• (3) Build your tools
• (4) Buy infra

Hiring

• (5) Hire agents before people
• (6) Build small and extremely high taste teams

Learning

• (7) Maximize internal information openness
• (8) Design a compounding learning machine

Process design

• (9) Develop rubrics
• (10) Reactive + Proactive

Task learning

• (11) Store and learn from production traces
• (12) Encode your expertise

(1) Own your context

Really, you have to own it. This means you need to avoid tools that want to hoard your context to create a moat. Either the vendor must give you complete and unlimited access to your data (and derived data) to use in any way you please, or you need a different solution. Your context is your explicit knowledge - but also all the tacit and institutional knowledge. For the first time you need to write all that down for the org to be legible to humans and agents.

(2) Rent your intelligence

Intelligence is not yet a commodity, but it will be for 99% of knowledge work. The way to maintain optionality and pricing power is to make it easy to swap intelligence. Don’t let the context vampires in, don’t let the fox into the hen house.

(3) Build your tools

Software is cheaper and easier than ever to make. This is shifting the line between buy and build. Your tooling stack is unique to your business and owning it fully is another way to defend your unique context.

(4) Buy your infrastructure

But you should not build everything. You (probably) should not build your own serverless training infrastructure or database or operating system. Infra primitives allow you to move faster, build more scalable and resilient systems, and, most importantly, build more unique context.

(5) Hire agents before people

This one is quite simple. for a new task, eg reviewing applicant resumes, first try to teach an agent to do it before hiring someone to do it.

(6) Build small and extremely high taste teams

Small, high-quality, teams can make decisions faster and with more trust. When the bottleneck moves from execution to taste - you must optimize for the new bottleneck.

(7) Maximize internal information openness

Organizations of the past siloed information and created fiefdoms. It's unlikely this ever worked particularly well, but it’s catastrophic for agents who are only as good as their context. Default to open for your small, high-quality, team.

(8) Design a compounding learning machine

The organizations that win will quite simply climb the hill that is serving their customer and market faster. or said differently- they will learn faster. you want to design your company to be a continual learning machine and reduce the cycle time as low as possible.

(9) Develop rubrics

For core and/or critical tasks you want to provide agents with ample instructions but also a scorecard to check their work. The scorecard or rubric, and their specific criteria, is also something to improve and iterate on over time.

(10) Reactive + Proactive

One temptation is to treat an agent like a lowly intern - you hand them the work and they execute it, hopefully well. This is fine but the colleagues who are truly a joy to work with are also proactive. They are thinking about the business, the customers, the roadmap constantly and proposing means to improve the company. Look for ways to build background agents, with sufficient organizational context, that can do this well.

(11) Store and learn from production traces

This one seems obvious but I think it’s more subtle than people realize. The goal is not to spend a bunch of money on telemetry “just in case” - but to use real execution of agents to help them improve. This isn’t quite “recursive self-improvement” - but for the applied AI builder is not far-off.

(12) Encode your expertise

The biggest gains to your agents will be identifying patterns of issues and correcting them. Other agents you employ can likely identify these patterns and then come to you to interview you so you can help fix them.

(13, bonus!) Most of all: think before you AI! Your thought and taste is, and will continue to be, your advantage.

Organizations that adopt these will be smaller, faster, and deliver outsized value to their customers. Organizations that do not will die. For more theory on this - read: The Quick shall inherit the Earth.

Good luck.

Thanks to @dexhorthy (author of 12 factor agents) for reading a draft of this article