Every few months a consultancy shows up with a deck titled Enterprise AI Transformation. 60 slides. A swim-lane diagram with fourteen agents. A twelve-month roadmap. A price tag somewhere between "new headcount" and "small acquisition."
And every few months I watch a mid-sized business sign it, spend six months in discovery, and ship nothing.
I sell the enterprise tier too. I charge $4,999 for it. And I'm going to spend the next thousand words telling you that most of you reading this shouldn't buy it yet — and explaining what you should buy instead, including from competitors.
This isn't reverse psychology. It's the only way enterprise AI actually works.
The transformation trap
The trap looks like this. A founder or ops lead reads a McKinsey article, goes to a board meeting, and says some version of: "We need an AI strategy." Three weeks later a vendor is quoting six figures for a cross-department rollout. Six months later there's a Notion doc full of architectural diagrams and exactly zero automations in production.
The failure mode is consistent. It isn't the vendor's fault. It isn't the tooling. It's a sequencing error.
Enterprise AI transformations assume you already know:
- Which workflows are actually painful (not just feel painful in a stand-up).
- Which of those workflows an LLM can reliably handle at your accuracy bar.
- What "good" output looks like well enough that you can write an eval for it.
- Who on your team will operate, monitor, and adjust the agents after the consultant leaves.
Most businesses do not know any of these things when they sign the contract. They can't — they've never run an agent in production. So the "strategy workshop" becomes an expensive discovery exercise, the build phase becomes a guessing game, and the handover is a pile of YAML no one on staff can read.
You cannot skip directly from zero agents to five coordinated agents across multiple departments. The distance isn't technical. It's organisational.
The real ladder
Here's the progression I've watched actually work, across something like thirty engagements now:
Rung 1: One human runs one agent for 30 days. A single person on your team — ideally the ops lead or the founder — picks one painful workflow and automates it. Not for the whole team. For themselves. They run it daily, notice where it breaks, learn what "good output" means in their own voice. Budget: a weekend and maybe $20 in API credits. Duration: one month minimum before progressing.
Skip this rung and everything downstream collapses. I'm not being dramatic. The person who understands how one agent fails is the only person who can sensibly design a system of five.
Rung 2: One agent across one department. Now take what that first person learned and deploy it to a small team. Three to five people using the same agent for the same workflow. This is where you discover the real infrastructure problems — access control, shared context, who fixes it when it hallucinates, who decides when to retrain the prompt. Budget: a week of someone's time plus ~$50-100/month in API costs. Duration: two months minimum.
Most businesses I talk to think they're ready for "enterprise transformation" when they're actually on rung 2, day 10. They haven't hit the second-week wall yet — the moment when the agent starts drifting, someone blames the AI, and leadership has to decide whether to invest in observability or retreat. Retreating is fine. Retreating and then signing a six-figure transformation contract is how you end up with the Notion doc and no agents.
Rung 3: Two or three agents, one department, coordinated. This is where orchestration starts to matter. You need one agent to hand work to another. You need logging. You need a human review step that isn't "check the Slack channel." You need someone on staff whose job description has "monitors AI systems" in it, even if that's 20% of their role. This is roughly my $649 single-system setup tier if you want help; it's a weekend of learning if you want to do it yourself. Duration: three months before the system stabilises.
Rung 4: Multi-department, shared infrastructure. Now you have agents in two or three parts of the business, they need to talk to each other, and you're tripping over the absence of a shared data layer. You need a proper orchestrator, cross-team runbooks, an on-call rotation (yes, really). This is my command-center tier at $2,199 — or roughly a senior engineer's month if you build it in-house. Duration: six months to steady state.
Rung 5: Enterprise transformation. Five-plus agents across multiple departments, custom development, team training, extended support. This is the $4,999 engagement on my site. It's real work, and I'll do it for you if you're ready. But I will push back — hard — if you try to buy it from rung 1.
The gap between rung 1 and rung 5 isn't twelve weeks of consulting. It's roughly nine to twelve months of organisational learning. You cannot compress that by paying more.
How to tell which rung you're actually on
Pick the first statement that isn't true for your business:
- Someone on your team has personally shipped an LLM automation they use daily.
- A team of 3+ people uses that automation and someone has fixed it when it broke.
- Two or more automations are running in production, with logs you can actually read.
- Automations exist in more than one department and they share data or context.
- You have a named person whose job includes monitoring and maintaining AI systems.
Whichever number is the first "no" — that's your rung. Start there. Don't buy the rung above it.
If statement 1 is a no: you need a weekend and a free tier OpenAI or Anthropic key, not a consultant. Every vendor in this space — including me — has a cheap or free starting product. Use any of them. I don't care which.
If statement 2 is a no but statement 1 is a yes: congratulations, you're the furthest ahead of anyone reading this. Keep running your thing solo for another month. Seriously. The impulse to "roll this out to the team" is almost always two months premature.
If you're a yes on 4 and a no on 5: now we're in enterprise-transformation territory. You've hit the ceiling of "one competent person can hold it all in their head." You need the org structure, the runbooks, the shared infrastructure, and the training. This is the rare case where a $5k engagement pays back inside a quarter.
Why I'm telling you not to buy
Because the alternative is worse for everyone.
A business that skips rungs and buys a transformation lands in one of three places:
- Six months of discovery that produces a deck and no production agents. Consultant gets paid, business gets nothing, industry gets another cautionary tale.
- A working system no one on staff understands. It runs for four months, the consultant's support window closes, it breaks, and it quietly gets turned off.
- A system that works but was five times more expensive than needed because the business bought the enterprise tier when it needed the single-system tier.
I don't want to be the person on the other end of any of those outcomes. My refund rate on the enterprise tier is zero precisely because I talk about 60% of inbound leads out of buying it. They come back six months later when they're actually on rung 4 and the engagement takes eight weeks instead of thirty.
What to do this week
If you're building toward enterprise AI and you don't know which rung you're on:
- Open a blank doc. Write down every repetitive task in your business that takes more than 20 minutes a day.
- Circle the three where the output is text, structured data, or a decision — not a physical action, not something requiring live human judgment.
- Pick one. Spend a weekend automating it for yourself.
- Run it for thirty days before showing it to anyone else.
When you come back in two months, you'll know exactly what you need to buy next. And it probably still won't be the enterprise tier. That's fine. The right time to buy it is when you already know what you'd build — you just don't have the weeks to build it.
Until then, stop reading decks. Go build rung 1.
Top comments (0)