Can Devin AI Replace Product Managers?

In the fall of last year, I wrote about whether AI could replace product managers. I tried to use ChatGPT for product management tasks.

As part of my conclusion, I wrote:

Devin transformed my opinion on this. Devin is an AI software engineer recently unveiled by Cognition Labs. When I say “software engineer,” I don’t mean another super-powered autocomplete that suggests code snippets and spots errors.

It’s hard to explain everything the model does. Just watch Scott Wu, the CEO of Cognition AI, present what Devin can do:

[Watch here]

According to Cognition Labs, Devin has already completed tasks on Upwork and passed technical coding tests. I heard the next version can even yell at StackOverflow users to use the search function!

It feels like a few months ago, the main AI narrative was that nobody would lose their job. Instead, we’d all just get AI superpowers that let us skip the boring stuff and do what matters.

Devin makes me think this was too optimistic. If Devin works that well on any given coding task (a genuine question, given that tech demos are usually presented under ideal circumstances), it obviates the need for at least a junior software engineer.

If we can build this for code, what else can we build it for? Product management is top of mind for me. My experiment last year was done purely in ChatGPT and leveraged none of the AI agent models or product-specific AI tools.

Could a “Devin for PMs” make product managers obsolete? Here’s why it might happen—and why not:

Many product management tasks can be done by LLMs: Writing tickets, user stories, spec docs, strategy memos, analyzing data, summarizing user feedback.

If you include agent models, AI could even research competitors or update projects and tickets. All of these can already be done.

All that’s missing is the ability to use Jira/Linear and other tools and to maintain specific knowledge across tools/domains. This might not be an easy problem to solve, but tools like Devin make me think it’s easier than it looks.

If you’ve been a PM for more than 10 minutes, you know that the user stories, product specs, and Jira tickets aren’t the most important part. Not even the strategic analysis (which AI agents could conceivably do).

The hardest (and most valuable) job of product managers is often taming leadership’s expectations and keeping engineers/designers motivated. It’s talking to users to discover their pain points. It’s being decisive and rallying a team behind those decisions to make sure they’re implemented.

At their core, these tasks are human relationship issues AI can’t tackle.

But models like Devin show us that product management will change. Here are some I could see happen:

Whether or not we ever see a “Devin for product management,” this model could affect PMs in a big way. How much would your job change if you had access to Devin?

Every product creates eng work most engineers hate. You can probably hear your eng team sigh just reading the words rearchitect, tech debt, or debugging. What if you could simply tell an AI to do those things?

Even though it does none of your product management tasks, Devin would change how you work. Now imagine adding “Devin for PMs” to that.

What if competitive research, writing spec docs, analyzing data, or updating tickets required a simple text prompt and 10 minutes of waiting?

Sure, none of this is reality yet. And I’m certain we’ll see technical, practical and regulatory roadblocks on the way to any of this becoming more than a sci-fi vision.

But we went from ChatGPT making up facts about radio towers in Lithuania to fully automated software engineers in less than two years. And AI shows no signs of slowing!

One company building AI tooling specifically for PMs is Yana Welinder's Kraftful (which describes itself as a copilot for product teams) : While it doesn't offer an autonomous agent like Devin, it automates much of the manual work product managers do—reading surveys, writing user stories, surfacing insights from interview transcripts...

If agent-based models like Devin become more common, Kraftful is in a perfect place to infuse its product with the power to execute the follow-on workflows: It already has all the training data it'd need and the API access the AI agent would navigate.

All of this makes me think that what happened to physical labor in the age of machines happens to knowledge work in the age of AI: We can get far more done with far smaller teams.

If the engineering and product teams we have today are like a workforce harvesting with scythes, AI is the combined harvest, allowing one person to harvest the entire field while listening to podcasts.

This article originally appeared on CommandBar.com.