AI Planning: Why the Perfect Plan Is Dead

You spent three weeks putting together the project document. By the time it was done, half of it was already wrong. If that sounds familiar, this post is for you. AI planning has become something else. Anyone still planning like it's 2019 is paying a lot for paper.

I'll explain where this shift comes from, what it means for people running a business, and where a plan is still worth its weight in gold.

Where the idea that "plan mode" is dead came from

There's a provocative line going around among programmers: "plan mode is dead."

A bit of context. AI tools that write code got a feature called plan mode. It works like this: before touching anything, the AI writes a detailed plan. You read it, approve it, and only then does it execute.

Sounds sensible. Measure twice, cut once.

The problem is that AI got too fast for that ritual to always make sense. Today it builds a first version in minutes. It's often cheaper to let it build, look at the result and fix it than to spend an hour discussing the plan.

The plan stopped being a document. It became the first working version.

This started in the coding world, but the lesson applies to any business.

Why the long plan became waste

The long plan had a reason to exist: mistakes were expensive. If building a system took six months, going in the wrong direction cost six months. So spending weeks thinking beforehand made sense.

That math has changed.

An automation prototype that used to take three weeks to get running now takes a day or two. A system screen that needed a designer, a meeting and an approval can be sketched in an afternoon.

When building gets cheap, the detailed plan loses value. It tries to predict answers you can simply test.

And there's a detail few people admit: long plans are almost always wrong. Not because of incompetence. You only find out what you need when you see the thing working. The client looks at the screen and says "oh, but I wanted the customer's tax ID to show up here." Nobody thought of that in the kickoff meeting.

The requirements document is the most well-loved product in the company: nobody reads it, so nobody complains.

An ugly prototype that runs teaches more than a beautiful document nobody tested.

AI planning in practice: the prototype is the plan

Here's a scenario I see often.

A service company wants to automate its quotes. Today an employee gets the request on WhatsApp, opens a spreadsheet, does the math, builds a PDF and sends it back. It takes about 40 minutes per quote.

The old way would look like this:

• Meeting to gather requirements.
• Document with every pricing rule.
• Document approval.
• Development.
• First delivery after a month or two.

The new way is different. Instead of writing down the rules, I take five real quotes that were already sent. I build a simple version that takes the request and generates the PDF. In two days the owner is looking at a quote that generated itself.

That's when things surface. "This discount only applies to returning customers." "When it's out of town, add the travel cost." Rules nobody would remember in a meeting, but that jump out when the result is right in front of you.

After three or four rounds like this, the automation runs smoothly. And the "plan" got written along the way, based on what actually worked.

Notice: planning still exists. It just changed shape. It left the paper and moved into the cycle of testing, looking and adjusting.

So you just start building? Where the plan still matters

No. And here's the part a lot of people excited about AI forget.

Testing fast works when mistakes are cheap. When mistakes are expensive, you plan. Simple as that.

Situations where I still stop and think hard before building:

• Real money moving. Automation that issues charges, touches accounts or applies discounts. A mistake there turns into a loss or an angry customer.
• Sensitive data. Customer information, health data, personal documents. Data protection laws don't accept "it was just a test."
• Integration with legacy systems. That ERP from 2008 nobody fully understands. Messing with it blindly can bring operations down.
• Things that are hard to undo. Sending a message to 5,000 customers. Deleting records. There's no "Ctrl+Z" for that.

In these cases, the plan isn't dead. It got more important, precisely because everything else got fast and people tend to rush where they shouldn't.

My rule is one question: if this goes wrong, what does it cost to roll back? If the answer is "an afternoon," I test. If it's "a lost client" or "a fine," I plan.

The most expensive mistake I see in small businesses

My opinion, straight up: a two-hour meeting to decide something you could test in twenty minutes is one of the worst uses of money in a company.

Do the math. Five people in a room, two hours, an average wage of $30 an hour. That's $300 to debate a hypothesis. With AI, someone builds a draft in half an hour and the discussion becomes about something concrete.

Discussing over a draft is much faster. People stop defending abstract ideas and start pointing at real things: "this field is unnecessary," "this button is confusing," "we never use this."

Another common mistake is the opposite: using AI to produce even bigger plans. I've seen people ask ChatGPT for a "complete digital transformation strategic plan" and get 30 beautiful pages. That ended up in a drawer. AI made it easy to generate documents, and a document nobody executes is still worth zero.

How to apply this in your business this week

If you want to try this way of working, start small. Here's a script that works:

1. Pick a boring, repetitive process. Something someone does every week that doesn't involve money or sensitive data. A weekly report, a standard customer reply, organizing a spreadsheet.
2. Gather real examples. Three to five cases of how the task is done today. A real example is worth more than any description.
3. Build the ugly version. Use an AI tool or get help building something that solves 70% of the problem. Don't aim for perfection.
4. Show it to the person who does the task. Whoever does it every day will find the gaps in five minutes.
5. Adjust and repeat. Two or three rounds are usually enough for something useful.
6. Only then document it. Write down how it turned out, so whoever comes next understands it.

Notice the order. The document comes at the end, describing what works. Not at the beginning, trying to guess.

What changes in a manager's mind

The big shift is emotional, not technical. A lot of people overplan because they're afraid of failing in public. A detailed plan gives a feeling of control.

But real control comes from seeing results early. A prototype on day two tells you more about the project's future than a well-designed timeline.

AI removed the cost of trying. Those who take advantage of it test ten ideas in the time a competitor approves one. Those who don't keep writing documents that go stale before they're finished.

If you have a process in your company stuck in "someday we'll automate it," maybe the first step is just a working draft, no 30-page plan. That's the kind of thing I do every day: design, automation and AI for businesses that want to get ideas off the ground without overcomplicating things. Get to know my work and let's talk.

You spent three weeks writing the project plan. By the time it was done, half of it was already wrong.

With AI, a working first version is ready in two days. Today the plan is the prototype, not the document.

The client looks at the screen and instantly remembers the rule nobody mentioned in the kickoff meeting.

But the plan isn't dead. When money, sensitive data or something you can't undo is involved, I stop and think before I build.

I ask one question: if this goes wrong, what does it cost to roll back? If it costs an afternoon, I test. If it costs a client, I plan.

Do you have a process stuck in "someday we'll automate it"? Tell me in the comments. Sometimes a working draft is all it takes.

#ArtificialIntelligence #Automation #ProjectManagement #Productivity #SmallBusiness