Product Managers in the Command Line (Interface)

For the second time in three years I’ve felt the world shift

Over the past year or so I’ve been spending an uncomfortable amount of time in the command line interface (CLI) for a non-coder. The command line, or terminal, is a text-based interface, a way of communicating to your computer mostly used by developers or those old enough to remember a time before Graphical User Interfaces (GUIs – the icons, windows, mouse movements we are all familiar with). It’s what you see hackers using in movies, all shortcuts and old-time fonts.

The terminal is powerful, fast, and a little intimidating at first but crossing this small technical barrier has outsized benefits. It feels like a developer space, it is mostly used for system administration, automation, scripting, installing, and file management.

Earlier in the year I was working in the terminal as my LLMs directed me to for my projects – I understood little and copy + pasted a lot. “Install this program” “Run this script” with me sending back any errors I encountered. With time & patience I was able to build real functionalities. API connections, Google Auths on my apps, and MCP servers for my self-hosted chatbot UI. It was by no means easy or efficient.

But things have changed. The arbitrary boundary is dissolving and non-developers can now work effectively from the CLI using plain English.

Claude Code

Now I am working in the terminal in a different way. Talking to ChatGPT those first few times a few years ago was like talking to the future, a whole world of new possibilities opening up. Using Claude Code was similar, a new (old!) way of working at a computer.

Claude Code brings Claude into the Command line on your machine. It can be given all the context on your computer. It has the power to make changes. It was an internal side project at Anthropic that grew into a billion dollar business in 6 months. For world class developers this is rapidly changing how they view their jobs.

Claude Code combines a harness & scaffolding with sets of rules and prompts that run the Claude LLMs – this helps them become great at tool calling and are especially good at agentic work. The newest Opus 4.5 model is being talked about as a watershed moment, a tipping point where software development moves from LLM copiloting a human, to the human performing Agent management and orchestration – not touching the code itself.

Boris Cherny, lead Claude Code developer:

It’s summoning a Ghost intelligence as Andrej Karpathy terms it, right in your context, right in your day to day.

What is happening for software development is coming for all computer work. Product Management, so close to software development, is in a prime position to be disrupted.

Claude Code for Product Managers

A big advantage is less context switching. No more copy + pasting between your LLM and your work – you can get the work added directly in your files or indeed have them created – right up to full apps.

A key ability is how you can prime Claude in each project and/or in general to follow your rules, your style with a CLAUDE.md file. This acts almost as a memory for Claude, key information giving it context that will be reread each time that Claude is in that particular space. This is where a map of the project code base should go, programming rules, but also for PMs little things like if you want it to always write in British English for a particular project. Add it here and Claude will follow it. Claude can write these files for you.

For PMs there are many areas where a typical workflow can be enhanced. As Claude Code has direct file access it understands your context, you don’t need to re-explain everything each time. You can get assistance in writing PRDs, in being questioned on your assumptions, and in real-time web-searched competitor analysis.

Claude Code can spin up sub-agents to do work in parallel – for instance if you have 10 user interviews to analyse, it can perform the same analysis ten times at once and synthesise the response into a well formatted document. You can write specific Agents, profiles that respond from a particular viewpoint such as Engineering or design. Claude can run complex data analysis complete with dashboards and visualisations without you ever writing a line of SQL yourself.

You can have it explore codebases with you, patiently explaining code in PM terms like the best engineering managers. You can have it spin up rapid, working prototypes that fit with the existing codebase. Claude Code can suggest and provide real automations for your workflows.

Claude Code is a true partner for me – in some ways Claude is kind of the senior partner. Claude’s writing the code locally, raising & pushing the PR in Github, syncing my site through Vercel. I just asked it to do something and gave it permission.

I had it activate the UI design “Skill” – a preset tool the agent can call and use – to one-shot a Product Portfolio site for me. It’s pretty good! On par with what I made with Lovable after a fair bit more prompting. This is of course a super basic use case – people are building full production businesses this way.

Caveats

For Claude Code to be effective it needs the context and permissions. On your home PC this is fine. But at work? Depends on the workplace. You may end up with lots of word files you need to copy & paste back to a cloud hosted app. Or, have MCP servers that can sync work to the right apps – like Figma or Linear, but again, this may run into security issues. It is not something you can do on your own.

While Anthropic has done a great job in softening up the UI with cute animations and clever wording, it will take a leap of faith into a more technical space for PMs to adopt widely. A softer version would spread like wildfire I am sure.

Competition

OpenAI have a rival tool, Codex, it was part of their app but is now also extensible into CLI and IDEs. It runs the Codex tuned 5.2 model that seems to outperform Opus on some tasks. It is not generating the same buzz online. Below I use it to read one of my code repos and explain to me what each file does.

Codex, like Google’s Jules, has many of the benefits of Claude Code, but the leap for non-developers feels larger – neither these or the other options feel like magic. This could change rapidly, I think there is fertile ground here in creating a dedicated user friendly tool with this power for non-developers. Potentially another billion dollar idea – but for me it’s already here, we just need to embrace Claude Code, the terminal is not just for hackers anymore!

Great in depth guide to Claude Code by Sankalp
My review of Claude Code for PMs course by Carl Vellotti