I’m beginning to agentify my workflows

I’m embracing AI, and I’m beginning to agentify most of my workflows.

“Agentify” is an awkward word, but it captures the direction I’m moving in. I want to look at the work I do regularly, understand the steps involved, and give an AI agent a useful part of that process to carry forward.

I’m still at the beginning. There is a lot to figure out: what to delegate, how much context to provide, where to put the checkpoints, and how to tell whether the result is actually better.

That is what makes this interesting to me. It is an opportunity to rethink how I work while building the habits to use these tools well.

Thinking in workflows

A prompt usually starts with a request: explain this, write that, help me solve a problem.

A workflow asks a wider set of questions. What starts the task? What information does it need? What should happen next? What does a useful result look like? Where should I step in?

Those questions are becoming the starting point for how I think about using AI.

A research workflow, for example, could move from gathering sources to organizing findings, surfacing disagreements, and preparing a brief for review. A software workflow could move from a clear requirement to an implementation, checks, and a summary of what changed.

These are examples of the shape I’m aiming for. Each step needs a purpose, and the handoffs need to be clear enough that I can see what happened.

Making the work explicit

There is a useful challenge hidden in this process: describing the work properly.

It is easy to carry a workflow around in your head. You know which sources you trust, what you mean by “finished,” and which details deserve another look. An agent needs those expectations spelled out.

What context should it read first? Which files can it change? What evidence should it bring back? Which decisions need my input?

Writing that down forces me to examine the process itself. A vague instruction often reveals an assumption I have never made explicit. A confusing handoff can point to a step that needs simplifying.

I want these instructions to become useful working documents that I can refine as I learn.

Giving agents room to work

For this to be useful, an agent needs enough room to make progress. At the same time, I want to be deliberate about what it can do and what I need to review.

Preparing a draft, investigating a problem, or proposing a code change can produce something concrete for me to inspect. Publishing a piece, sending a message, or making a consequential change deserves a clear checkpoint.

The level of independence should fit the task. I want to start with work where the result is easy to examine and mistakes are easy to correct, then expand as the process earns my confidence.

Keeping a record of the instructions, actions, and results should help me understand where an agent is useful and where the workflow needs more attention.

Staying responsible for the result

I still need to understand what I’m asking for and decide whether the output is good enough.

For research, that means checking sources and noticing where an answer is uncertain. For writing, it means making sure the piece says what I mean. For software, it means reviewing the changes and checking the behavior.

I want verification to be part of the workflow from the beginning. An agent should bring back something I can assess: the relevant sources, the reasoning behind a choice, or the checks it actually performed.

That gives me a better basis for the next decision than a confident statement that the task is complete.

Beginning, then improving

My intention is to keep moving more of my everyday work into this model, one workflow at a time.

Some processes will be straightforward. Others will need several attempts before the instructions and checkpoints feel right. I expect to revise them as I discover what works for me.

The practical question is simple: does this help me move useful work forward, with a result I can understand and stand behind?

That is how I want to approach embracing AI. Start with real work, make the process clear, review the outcome, and keep improving it.