AI Will Take Longer Than I Thought

I have to admit that I have become more frustrated with AI over the last two or three months, and less certain that it will spread through the economy as quickly as most people predicted.

That includes me. I was too optimistic.

The strange thing is that the technology itself isn’t disappointing me. AI has become incredibly useful in my own work.

I’ve become a huge fan of OpenAI’s Codex. I use it as the interface for an AI chief of staff I built called Tiro.

Tiro helps me triage four email accounts, remember the companies and projects I’m working on, organize travel, analyze financial statements, track workouts and meals, maintain relationships, and think through strategic and marketing decisions. Different specialists keep their own memories, files and dashboards, but I can work with all of them by talking through Codex.

It is imperfect, but it is real. I use it every day.

That experience has convinced me that one motivated person can already build enormously useful AI systems for himself.

But my work with our startup, Rustproof, has shown me why building companywide AI systems is so difficult.

It is true that you don’t need traditional software developers for everything anymore.

But you do need someone who understands the business, is willing to learn the basics of software, can operate tools like Codex, and is sufficiently curious and self-directed to keep working through all the inevitable problems.

Those people exist inside companies. They are usually some of the company’s best people.

That creates a problem. To build a serious internal AI system, you effectively have to take one, two or three of those people offline and let them go deep for months. They have to map the process, connect the data, test the system, correct it and understand why it failed.

I’ve been deep in this work for months and it keeps piling up. We are constantly having to build internal tools that should be plug and play products/services we buy from someone else. They just don’t exist yet.

Most established companies won’t or can’t make that investment. Their best people already have jobs.

Alternatively, companies can hire someone from outside, but that presents a different problem. Most development firms still operate like traditional software companies. They gather requirements, write code, review every line, run a project and charge accordingly.

That process is too expensive for most small and midsized businesses. It is also poorly suited to AI systems that have to be shaped through constant use and correction.

The kind of company we need is part software builder, part workflow designer, part operator and part change-management partner. It needs to build the system, run it, monitor it and keep improving it as the business changes. Rustproof is exactly that company but we focus on a very narrow niche.

Most other niches don’t have that company yet or at least at a price SMBs can afford.

I expect most existing companies will continue to struggle to transform themselves quickly by adding AI to everything they already do. They have too many systems, processes, permissions, incentives and people to coordinate.

Instead, AI may enter many industries through new competitors.

A founder will build an agent to solve a problem for himself. He will gradually add memory, data, controls and additional workflows. Eventually, what began as a personal tool will become the operating system of a new business.

That company won’t have to retrofit AI into an existing organization. It will be built around AI from the beginning.

It will deliver the same service with fewer people, faster turnaround and a radically different cost structure.

By the time established competitors recognize what happened, they won’t be competing against an AI product. They’ll be competing against an AI-native company.

There is bad news and good news for existing business owners.

The bad news is that these AI-native competitors are coming.

The good news is that it will take them time to appear, earn customers’ trust and climb the quality ladder far enough to become serious competitors.

That creates an opening.

AI is already powerful enough to change the economy, but entrepreneurs have not yet built the companies needed to carry it into ordinary businesses.

Traditional consultants can tell an owner what to do. Traditional software developers can build what the owner specifies. Neither model is quite right for AI.

What the owner really needs is someone who can understand the business, identify a valuable workflow, build the agent, connect it to the company’s information, monitor what it does and keep improving it after launch.

This is less like selling software and more like providing a managed service powered by software. The customer shouldn’t have to become an expert in agents, models, databases or observability. The provider should absorb that complexity and deliver a dependable business outcome.

That is a company someone can start today.

You might begin with one industry and one recurring problem. Build the workflow for one customer. Operate it manually where necessary. Learn what breaks. Turn the repeated pieces into software. Then use that system to serve the next customer more efficiently.

Over time, the service becomes a product and the product becomes an operating system for that corner of the industry.

Every owner should learn as much as possible about AI.

It will create immediate personal gains. It will make you a more informed buyer. It will also help you recognize where your business is vulnerable to a competitor built around an entirely different cost structure.

But the larger opportunity may belong to the entrepreneurs willing to build the missing layer between today’s powerful AI models and the messy reality of an operating business.

We need firms that can build, operate and improve AI systems for existing owners at a price they can afford.

Those companies have not been founded yet, at least not in sufficient numbers.

Start building them before the kids from Y Combinator arrive and simply take the markets those owners spent decades serving.