← Back

There Are Four Stages of AI Adoption. Most Businesses Are Stuck at Stage Two.

·9 min read·AI & Automation

Written by Derek Chua, digital marketing consultant and founder of Magnified Technologies. He runs a multi-agent AI system in production for content strategy, SEO, and client work.

Last week I came across a talk by Simon Willison, one of the most honest voices in practical AI engineering. He described how software developers move through distinct stages as they adopt AI tools. The moment I heard it, I stopped thinking about code entirely.

Because the same four stages apply to every role in every business. Marketing. Finance. Customer service. Operations. The pattern is identical, whether you are running a software team in San Francisco or an SME in Singapore.

Key Takeaway: AI adoption follows a predictable four-stage progression. Most businesses are stuck at Stage 2 — using AI to draft fragments — and haven't crossed the inflection point where AI actually takes over complete workflows.

Here is the framework, translated for business leaders.

Stage 1: AI as a Very Smart Search Engine

This is where most people start. You have ChatGPT, Claude, or Gemini open in a browser tab. You ask it questions. Sometimes it gives you a useful answer. Sometimes it hallucinates and you catch it.

You are essentially using AI the same way you used Google, except it can synthesise information and write in complete sentences. Useful, but not transformative.

Signs you are in Stage 1:

  • You open the AI tool when you have a question, but close it immediately after
  • You paste things in and read the output, then mostly rewrite it yourself
  • You do not trust the output enough to use it without significant editing

Most non-technical users hit Stage 1 in 2022 and 2023. A lot of them are still there.

Stage 2: AI Handles the Drafting, You Handle the Thinking

This is where most businesses are today. AI writes the first draft. You review it, reshape it, and make decisions. The AI handles mechanical work; you handle judgment.

This is genuinely useful. It saves hours on email drafts, reports, proposals, social posts, and internal documentation. But it has a ceiling: you are still the throughput constraint. Every output passes through your review before anything happens.

Signs you are in Stage 2:

  • AI drafts emails, documents, or content that you then edit heavily
  • You would not publish or send anything without reading the whole thing first
  • The AI saves you time, but your bottleneck is still you

This is where most businesses sit in 2025 and early 2026. Not bad. But also not close to the potential.

Stage 3: The Inflection Point

This is the one that changes things.

Simon Willison described it this way for developers: there is a moment when the AI writes more code than you do. That moment is the inflection point. Something shifts in how you relate to the tool and what you believe it can handle.

For business roles, it looks like this:

  • Your AI drafts an email, you read the subject line and the first two sentences, and you hit send
  • A workflow produces a weekly report that goes to the client directly
  • An agent processes inbound enquiries and you review the ones it flags, not all of them

The key distinction from Stage 2: you are no longer reviewing everything. You are spot-checking. You have developed a calibrated sense of what the AI handles reliably and what still needs your eyes.

This requires something that Stage 2 does not: you have to have tested the AI enough to know where it fails. Trust is not given; it is earned through observation.

In my own experience at Magnified, the inflection point for content happened when I let the SEO agent publish articles automatically at a certain quality threshold. I had reviewed enough of its output to know that when it scored itself above 8 out of 10 on the quality rubric, the article was reliably good. Now I mostly check what it decides not to publish, not what it does.

Stage 4: You Trust the Process, Not the Output

This is where things get genuinely uncomfortable for most people.

Simon mentioned a company called StrongDM that described their approach as: nobody writes any code, nobody reads any code. He called it "wildly irresponsible" — and he was right. That extreme version is reckless. But the underlying shift it points to is real.

Stage 4 is not "AI runs everything unsupervised." It is a different kind of oversight. Instead of reviewing individual outputs, you trust the system. You build checks into the process itself. You monitor for failures at the pattern level rather than the line-item level.

A legal team at Stage 4 does not read every clause the AI drafted. They review contracts above a certain size, flag specific clause types for human eyes, and audit outcomes quarterly.

A customer service team at Stage 4 does not read every reply the AI generated. They monitor resolution rates, scan for complaints, and have escalation rules that bring human judgment in automatically.

The question is not whether the AI is good enough to run workflows without you reading every output. The question is whether your process is good enough to catch failures reliably.

Most businesses are nowhere near ready for Stage 4. And that is fine. But understanding it exists changes what you build toward.

Where Most Businesses Actually Are

At Magnified, the pattern I see across clients is consistent:

  • Small businesses: mostly Stage 1, occasionally Stage 2
  • Growing SMEs with a designated AI champion: usually mid-Stage 2
  • Companies that have invested in AI tools and training: approaching Stage 3 for one or two workflows
  • Very few: operating anywhere near Stage 4 for any process

The gap between Stage 2 and Stage 3 is not a technology gap. The models are capable of Stage 3 work today. The gap is a trust gap, and it only closes one way: by running experiments, observing failures, and building calibrated judgment about where the AI earns autonomy and where it does not.

The Honest Question to Ask Yourself

For each major workflow in your business, ask: which stage are we at, and why?

If the answer to "why" is "because we haven't tried going further," that is not a risk decision. That is inertia. The risk calculation is different if the answer is "because we tested it and the failure rate was unacceptable."

Willison's most useful observation was about tests in software development. Tests used to be skipped because writing them was extra work. Now that AI writes tests automatically, they are effectively free. So skipping them is no longer a time-saving decision; it is just a quality decision.

The same logic applies in business. The work required to run a trial of a more autonomous AI workflow is much lower than it used to be. If you have not attempted Stage 3 because it seems like a lot of effort, that excuse has a much shorter shelf life than it did a year ago.

At Magnified, We Have Seen This Pattern

The businesses that make meaningful progress with AI in 2026 are not the ones that adopt the most tools. They are the ones that move deliberately through the stages, build confidence at each level, and make intentional decisions about where to extend trust.

The ones that plateau are typically stuck in Stage 2 indefinitely because they treat every AI output as something to be reviewed completely before use. That approach does not scale. It also does not produce the efficiency gains that make AI adoption economically compelling.

Pick one workflow. Move it to Stage 3. See what you learn.

If you want help working through which workflows in your business are ready for Stage 3, get in touch with Magnified. That's exactly the kind of structured AI adoption work we do with clients.

Frequently Asked Questions

What are the four stages of AI adoption for businesses? Stage 1 is using AI as a smart search engine to answer questions. Stage 2 is using AI to draft content or documents that you then edit. Stage 3 is the inflection point where AI handles complete workflows and you spot-check rather than review everything. Stage 4 is trusting the process rather than individual outputs, with systemic checks rather than manual review.

How do you know when it is safe to move from Stage 2 to Stage 3? You need enough experience with the tool's outputs to know where it reliably succeeds and where it fails. The best way to build that confidence is to run a controlled trial: have the AI complete a workflow fully, review its output as you normally would, and track where you made changes and why. After 20-30 cycles, you will have a clear picture of what you can trust.

Does Stage 4 mean AI runs everything without humans? No. Stage 4 means oversight shifts from reviewing individual outputs to monitoring the process and its outcomes. You build checks into the system, set escalation rules for edge cases, and audit results periodically. Human judgment is still in the loop; it just operates at the system level, not the line-item level.

Is it realistic for a small business to reach Stage 3? Yes, for specific workflows. You do not need to move the whole business to Stage 3 at once. Pick one repeatable, low-stakes workflow, such as social media drafts, first-pass customer responses, or internal status reports, and move that workflow to Stage 3. The learning from one workflow will accelerate progress on the next.

← All posts