It’s an open secret which people will be replaced by AI and which will thrive because of it. It’s true that you need to learn how to use AI effectively, but that’s not what I mean.

No, the people that will benefit from AI are the doers.

Not everyone is a doer. I see many information workers (people that sit in front of computers all day) get caught in three traps of not-doing:

• Business Processors
• Data agonizers
• Framework finders

This doesn’t have to be you! You don’t need to be one that has to reskill and move into a blue collar job. Read on about how to use AI to pull yourself out of these traps and become a doer that is set to thrive for years to come.

Business Processors

The processes of a business are often necessary. In software development, vibe coding new features and yeeting them into production without running tests is a bad idea. If business processes are valuable, it clearly follows that improving a business process is valuable.

When business process becomes a trap

And then there’s the next step that too many people take: if improving a business process is valuable, then adding a new process should be valuable too. This is not true, and it’s a serious trap.

The value was never in the process, but in the outcome the process achieved! If a new process doesn’t improve outcomes, it is a waste of everyone’s time. Worse, if you let “improving business processes” turn into your skill, brand, and mission, you are wasting all of your time. The real goal should be to remove every business process while still getting the outcomes.

That’s exactly what AI is good at. Agents powered by LLMs are extremely effective at business processses. When set up appropriately, they can reduce business process dramatically while keeping the outcome. For example:

• Businesses can improve inventory processes with image recognition, data entry, and even supplier coordination.
• You can convert project status reports the task tracking system into every report necessary
• AI can monitor tax regulations communicate the impact to account clerks

Is AI the end of process management?

It shouldn’t be a surprise that people with rote jobs are going to have less work in the future. I think there is a stronger lesson as well: managers of these processes will also soon stop delivering value. When AI that is flexible to discrepancies and changes can achieve the outcome, the job of the manager to adjust and oversee processes is less important also.

What that means is that anyone that delivers value by creating and managing business processes are about to stop delivering value. That’s not a good place to be.

For now, you can create a lot of value by implementing that process change to use AI. But once you’re done, that’s it. You’ve automated yourself out of a job.

Data Agonizers

Another trap, perhaps especially common in software development, is waiting to make a decision until you have all the data. In Microsoft 365 Copilot last year, we had to make a difficult decision about whether to continue with implicit use of plugins as our extensibility strategy (i.e. the AI decides whether to use an API across all APIs available). Here’s how this could have gone:

1. Partners start reporting that their plugins are not working very well
2. We look at the data we have and realize we can’t tell just from DSATs filed and service telemetry whether a plugin was supposed to be used
3. We “urgently” schedule a work item for next month to add telemetry for whether a prompt used a word in the plugin’s name
4. There’s nothing to do until the new telemetry is checked in, and that the telemetry code runs for a couple weeks to generate data
5. Six weeks later, we realize that the data is extremely noisy from partners testing their in-development plugins. This is messing up our analysis, so we log a bug to update the telemetry.
6. We wait for the next month plus two weeks
7. Repeat this cycle a few more times

In 4-6 months we could have gotten a definitive answer.

The cost of waiting for perfect data

Of course, we’d still have questions after those six months. Perhaps we would want to know whether some focused data science on the problem would improve results.

We did not take 6 months to make this decision. Instead, we saw the problem, analyzed the limited data that we had, used our intuition that the likely causes would not be easily resolved, and we moved on (to explicit use of agents). It was a tough decision, but it was the right one.

The trap isn’t analyzing data, but refusing to make a decision until the data proves it beyond a shadow of a doubt. The AI software industry moves too quickly; if we hesitatte, we will fall behind. Instead, use the best information you have to quickly make a decision.

AI will analyze better than you anyway

It’s also true that AI is about to be extremely good at data analysis. Microsoft’s Analyst agent can complete tasks that I’m not sure I ever could. Traiing AI to analyze data is nearly as easy as training it to program. When you need to analyze your readily available data in the future, an AI is going to do it. You won’t be able to rely on data analysis alone for adding value.

Framework Finders

The third trap of not-doing is relying on frameworks. When I joined Microsoft, I was surprised how little structure we had to development and processes. The product had overall milestone dates, but every team was left to do their own management. I had been told in college that “The best software companies, like Microsoft, are at level 5 of the software development Capability Maturity Model.” That was very far from the case!

In actual fact, Microsoft has learned that every product and team is different. A framework that works well for one team this year will not work well for a different team next year. Instead of learning, scaling out, applying, and optimizing, individual teams regularly reform their processes.

I like this way of executing much better.

When frameworks slow you down, not speed you up

Frameworks like SWOT or Five Forces analysis have their place: a quick way to make sure you’re not missing anything big. They’re good for about one discussion, and then my opinion is that they should be discarded.

The problem is that over-reliance on a framework stifles your thinking. The business school or consulting firm that developed the system were not in the situation you are in right now.

Any time that decisions and strategy can be this formulaic, guess what, AI can do it better than a human. Use an framework briefly—ideally directly with an AI instead of taking other peoples’ time—and then move on.

Characteristics of Great PMs

What do rigid business processes, waiting for exhaustive data, and overuse of frameworks have in common? They’re mechanisms of passing the buck on making a decision. There is risk when we must make a decision, and that doesn’t feel good. It’s easy to fall into the trap of wanting someone or something else to make the decision for you.

If you want to succeed in the world of AI, you must step up. Solve the problem in front of you as best as you can. Don’t fall back on process. You’ll never have perfect data. Don’t use frameworks as a way to slow down discussions until they are exhaustive. Make decisions.

Do.

We used to call this mindset of making decisions being a “self-starter” or having grit. It’s popular now to call it “high agency.” We shouldn’t be surprised that having a lot of agency is valuable in combination with AI. Computers have never operated with high agency. Even AI “agents” are called that only because they have the tiniest amount of agency more than traditional software.

We train AI agents to follow instructions, not make decisions. They won’t catch up to our ability to decide and own up to the resulting success or failure. Agents cannot make decisions outside of narrow domains.

Being a doer is our durable advantage for years to come. AI will not replace doers; it will supercharge them.