The Spring 2026 YCombinator “Requests for Startups” included a particular prompt I giggled about for a second: “Cursor for Product Managers”. After this request was published, a former co-worker approached me with an interview request about how I, as a product manager, use agentic coding. I agreed, and, after I answered some questions, he revealed that he was doing customer interviews for this YC request.

With a heavy heart, I had to tell him that at the company where I worked, Cursor already was the “Cursor for Product Managers”, and Claude Code was the Claude Code for product managers. If anything, this YC request read to me as an attempt to grab some quick cash before PMs shake off their omnipresent impostor syndrome and realize they can use the same general-purpose agentic coding tools the engineers use.

With the introduction of agentic AI and the hard skill barrier between the software engineer and the product manager evaporating, I see a lot of canonically “non-technical people”, like Product Managers, Sales Managers, or even Customer Support specialists trying agentic development. One problem I see with a lot of Product Managers specifically is that they’re trying to build the “perfect stack” of specialized apps.

There’s a belief in the community that once you settle on a particular set of tools, it will magically make your workdays easier and improve your output tenfold. In this article, I want to make the case for not fixating on tools as much, and focusing instead on the skills and context that make any tool work.

The “specialized AI tool for Product Managers” is a marketing strategy

It’s very interesting to me that quite a few people believe that the present-day agentic development tools are somehow too complicated to use or don’t fit the Product Manager skillset, while it’s exactly the opposite. I recently hosted an entry-level “vibecoding” workshop in Novi Sad, Serbia, and the feedback from the mostly non-technical participants was exactly that: “I didn’t believe agentic development is, in fact, that easy! You just talk to the model and learn as you go”.

Indeed, you can just talk to the model and get the results you want — and, as PMs, we’re uniquely equipped to know how to talk to the model to get results. It’s basically the same as giving a Jira task to a junior developer who is technically stellar, but needs a lot of guidance. The PM skill that matters most now is the same one you use reviewing a junior's work: knowing when output is confidently wrong.

Obviously, there are legitimate cases for specialized tooling. Heavily regulated environments — healthcare, finance, enterprise procurement — genuinely need more safeguarding than a standard agentic coding tool provides. But I’m talking about tools “specialized” for Product Managers, and PM work is, at its core, general-purpose knowledge work — writing, querying, prototyping, synthesizing. This is precisely what general agentic tools already do. There's no capability gap for a specialized tool to fill.

Tools can’t compensate for lack of direction

There are now numerous memes about people becoming slop cannons and getting too overjoyed with the ability to produce a lot of content quickly. So now it’s more important than ever to develop the soft skills — discernment, assertion, the proverbial “good taste”.

Under the buzzwords, it all boils down to this: you need to know what to do, what NOT to do, and how to present it to others. These skills are tool-agnostic. If you can describe your feature idea on a napkin, you will be able to present it better with an interactive prototype your agent made. But it still requires direction and a clear understanding of the goal you want to achieve.

We all have coworkers who have discovered Gen AI and now send a dozen 10-pagers for other people to “review” weekly. This is a soft skill issue, not a tool misuse issue. Tools come and go. Today we have Codex and Claude Code, tomorrow we may have something else. But your clarity of thought, your discernment, and your review instincts stay with you forever.

The most important prerequisite for the AI tools to work is context

With AI agentic development and tasks that don’t require writing code (like building PRDs, creating strategy decks, or working with customer feedback), context is king, and to build that context you need access to data: previous PRDs, product codebase, datamarts, customer interview data, meeting notes, Slack conversations.

Some companies will straight up allow connecting the AI tools to the company resources using MCPs to build that context. Some won’t, for compliance reasons, but may allow you to use locally-run models with content you gathered on your machine.

My point here is that it’s good to have access to frontier models, but for the Product Manager the value doesn’t lie in accessing the models that scored the highest on general knowledge or SWE benchmarks. Instead, it lies in the ability to build enough valuable context to outsource parts of the intellectual work. Even the best tool will produce a generic result if it doesn’t have any data on the product you work on, your customers’ or stakeholders’ pain points, and your code and infrastructure limitations.

As an AI-forward Product Manager, your goal should not be choosing the best tool, but building the best context for the tool to pull from instead.

My humble AI-powered Product Manager stack

Now I will expose my own “stack” that saves me hours weekly in both my job and my personal projects. It’s not much, but it’s honest work.

Something to code with: Claude Code

Hands down, my most used agentic development app. But, as I’ve said before, the power here lies in the context I built for my personal and professional tasks, with connectors to almost every tool I use daily, and the custom skills and routines I built for the day-to-day tasks.

Here's one example. Like every manager, I make small promises all day — "I'll check with the team and get back to you", "let me pull that number" — and they scatter across a dozen threads and meeting recordings. So I built a routine that pulls my recent messages and my meeting transcripts from Granola via the built-in connectors, and extracts every commitment I made: what I promised, to whom, and by when (if I was foolish enough to name a date). The output lands as a markdown checklist in my Obsidian vault, so it lives next to the rest of my notes. The whole thing took under an hour to set up, most of which was me describing what counts as a "commitment" versus me just being polite in a thread. It runs daily at the end of my workday as a routine job.

Something to publish the work: Vercel

Again, being very basic here, but my go-to is the most popular app. It hosts my personal website, it hosts my workshop handouts and pet projects, and so far I’ve only spent $20 on it to purchase the domain I use, so their free tier is very generous. I also tried Coolify (an open-source Vercel counterpart) and Google Apps Script, and, frankly, it’s all the same if it does the job — but Vercel and Coolify have better deployment flows.

Something to store knowledge on: Obsidian

Even though Obsidian sometimes feels like a glorified Notes app, it’s good to use with AI because its file system is very accessible to the agents, and anything I discover or want to put down or go back to can be stored there in a very organized way.

That’s it: three tools, some custom workflows, and a lot of context. As a Product Manager, what AI-powered workflow improvement has paid off most for you?