stormlamp.
The way through

Five stages of AI, and the one you can skip.

Every business is in the same storm. AI is changing how work gets done, all at once, and nobody got to pick the timing. So businesses reach for AI themselves, and it helps. Yet the damage most of them take doesn’t come from the storm. It comes from their own AI, spreading through the work until nobody can say what it’s doing. That part you can skip.

We’ll follow one everyday job through all five stages: Maria in sales, turning a customer’s request into a quote.

The story is an illustration, not a client result.

Stage 1

People ask AI

AI answers questions. People run every step.

A few people on your team use ChatGPT or Claude to clean up emails or summarize documents. It helps, but the AI never touches the work itself. If nobody uses AI yet, you’re here too.

The quote

Maria has ChatGPT polish her email. She still looks up prices in the spreadsheet, calls the warehouse about stock, and waits on Tom in finance to sign off on the margin. It takes two days, same as last year.

What it feels like

“We’re using AI.” The work itself runs the same way.

Stage 2

People hand AI tasks

AI does real work, set up by each person their own way.

Maria signs up for an AI assistant on her own card, connects it to her email and the shared drive, and it starts drafting her quotes. A colleague does the same for reorders with a different tool. Ask around: it may already be happening at your company.

The quote

Maria’s assistant drafts quotes in minutes from the price spreadsheet she pointed it at in the spring. Customers hear back the same morning instead of two days later. Maria reads every draft before she sends it.

What it feels like

Fast and impressive, and it really works. Each setup knows only what its person told it.

Stage 3 · Avoidable

The work gets away from you

AI is all through the work, and nobody can say what it’s doing.

Setups multiply. Assistants draft quotes, reorders, and follow-ups, each set up a little differently, some passing work to each other. Each one saves time on its own. Together, they become part of how the business runs, without anyone owning the whole of it. Each was easy to start. Turning one off now means first finding out what depends on it.

The usual fix is more AI: one shared company assistant for everyone, or another assistant to check the first. It cuts the clutter and adds a new way to fail.

The quote

In June, prices go up in the ERP, and nothing updates Maria’s assistant. For three weeks, quotes go out below cost, faster than ever. Nothing flags it until the margin report.

The shared fix: Some companies clean this up by giving everyone one shared company AI. Now one change hits everyone at once. Someone tweaks an instruction to fix the wording for one customer, and that week every quote leaves off the freight charge.

Who knows how it works: If Maria leaves, the way your quotes get done leaves with her: the instructions, the price sheet, and the history in her inbox.

What it feels like

Busier than ever, and less sure what’s going out the door.

The turn

The fix isn’t less AI.It’s the same fire, in a lamp.

Everything AI was already doing in the work, brought into one process your business owns: built on the systems you already run, tested against real work, and watched as it runs. Then you can see what it’s doing, and take it with you into whatever comes next. That takes building, whether by you, by someone you hire, or by us.

You don’t have to go through Stage 3 to get here. Coming straight from Stage 1 or 2 is the easiest route, because there’s less to untangle first.

Start a conversation

Not sure what’s running? Finding out is what the one-week Workflow Diagnostic is for.

Stage 4

You can see what’s running

You see what’s running. Then the quote becomes one process you can check.

It starts with finding out what’s actually running: every assistant and automation that touches the quote, what each one reads, what it costs, and where it gets things wrong.

Then the quote becomes one process, built on the systems you already have: your ERP, your price lists, your inbox. Plain rules handle what rules can, and AI handles the parts that need judgment, like reading a messy request. Every quote is checked against your rules before it goes out. Is this today’s price? Is the margin above your floor? Is freight included?

The quote

A request comes in. The system pulls today’s prices and stock, applies your margin rules, and drafts the quote. Maria approves every quote before it goes out, and anything unusual, like a new customer, a big discount, or an odd part, is flagged for a closer look. When prices change, they change in one place, and every quote uses the new prices.

What it takes: Most companies’ data is messier than they’d like. Part of the work is fixing the gaps the quote depends on, like item codes that don’t match or prices that live in three different places.

Tests from real work: Real examples, including the awkward ones, become repeatable tests, called evaluations, or evals. Every change to the quote has to pass them before it goes live.

See what is happening: You can trace a quote back to the prices it used, the checks it passed, and the person who approved it. You can see where work stalled and what happened afterward. We monitor errors and changes in performance; affected work can return to human review.

What it feels like

You know what’s running, what it costs, and what it got wrong. Prices change in one place.

The same quote, before and after the turn.

Switch between them to see where the quote waits, and what gets checked before it goes out.

Every step waits on a person.

  1. Request arrives

    Maria

    Maria reads the email and works out what the customer needs.

  2. Prices and stock

    Maria, warehouse

    Maria looks up prices in the spreadsheet and calls the warehouse about stock.

  3. Draft and checks

    Maria

    Maria builds the quote by hand and checks the math herself.

  4. Approval

    Maria, Tom

    Maria waits on Tom in finance to sign off on the margin.

  5. Records and follow-up

    Maria

    Maria updates the tracker and remembers to follow up.

Stage 5

Ready for what’s next

The business grows with AI, one tested decision at a time.

The storm doesn’t stop: AI keeps changing every month. At this stage, that’s good news. The system takes on work the way a new hire earns trust, one job at a time, each proved first.

It starts with routine quotes. Could the system send some itself? First, it has to show which ones it can handle and when it should ask Maria. We test ordinary cases, difficult cases, and past mistakes against outcomes your team has agreed are acceptable.

The quote

The requested part is out of stock. The approved substitute costs more, so the margin comes in under your usual target. That used to hold Maria’s quote until Tom signed off. Tom has written down when he’ll accept a thinner margin. The system was tested against his past calls, then ran beside him in shadow mode before it could decide. Now it makes the call within his policy, never past your hard limits on price or margin, and sends the quote. Anything outside it goes to Tom, with the options and the reason.

Shadow mode: Before it acts, the system runs beside the person who owns the decision. It recommends “send” or “ask Maria,” but cannot send anything, and we compare its recommendations with reviewed decisions on real work. Did it miss a risky discount? Did it ask for help on a routine order? This is shadow mode: learning whether it is ready, without handing over control.

One decision at a time: Doing one job well doesn’t earn the next. Each decision gets its own tests, its own run in shadow mode, and its own monitoring once it’s live. AI can judge when a request needs a person, but it cannot waive your hard limits on price or margin, and we test that decision to ask for help, as well as the quote it produces. Permission comes from those results, not from how sure the system sounds, and it covers that decision only.

When better AI arrives: Because every decision already has its tests and is watched as it runs, a better model doesn’t mean starting over. It runs against the same tests before it replaces what’s running, and if results slip after a change, that decision can go back to a person while it’s fixed.

Your people: Tom writes the policy instead of signing off on every exception. The people who used to make those calls set the limits and handle what’s genuinely new. Every decision is visible, none depends on one person remembering to check it, and you can take any of it back.

What it feels like

A better model is an opening, not a threat. It has to pass the same tests, and then it goes to work for you.

Quotes are just the example. The same stages apply to any work that starts with something arriving and ends with something done: lead to first reply, job to invoice, invoice to payment, inquiry to scheduled job, order to delivery.

See how we work, and what it costs
Start a conversation

Start with one workflow.

Tell us what happens today, where it gets stuck, and what you’d like to change. We reply within two working days.

Direct inquiry

hello@stormlamp.ai

A sentence or two is enough.