We use AI more and more in our daily work, and there is no sign of this slowing down. Writing, research, analysis, prototyping, summarising calls, checking requirements, preparing strategy docs, testing product ideas - all of it can already be supported by AI.
But there is an important choice hiding underneath the hype.
Right now, there are two broad paths a PM can follow: use a cloud LLM, or run a local one.
At first, this sounds like a technical choice. It is not. It is a career choice.
Do PMs need AI?
First, let’s be honest: you do not have to use an LLM.
The internet creates a very predictable FOMO loop: use AI or become irrelevant. Automate everything or fall behind. Become a 100x PM or be replaced by one.
The real world is more nuanced.
Some PMs will benefit massively from AI. Some will not. Some organisations are ready for it. Some are not. Some problems are worth automating. Some are better solved by talking to people, changing incentives, or removing unnecessary process.
As a PM, the real skill is not “using AI”.
The real skill is understanding whether AI helps with the actual problems you are solving.
What work do you repeatedly do? Where do you lose time? Which decisions require synthesis from messy inputs? Where do you need speed, not perfection? Where could a good first draft, research pass, or structured analysis make you better?
That is where AI starts to matter.
And if the answer is yes, if you do find meaningful use cases, then the next question becomes more interesting.
Not “should I use Claude or ChatGPT?”
But: should you use a cloud LLM or a local one?
Cloud LLM vs local LLM
A cloud LLM is what most people use today: ChatGPT, Claude, Gemini, Grok and similar tools. You can use them through a web interface or through an API.A local LLM is a model you run on your own machine. It could be one of the open-source models, with or without an interface wrapper such as OpenWebUI or LM Studio.
Cloud LLMs are easy. You open the app and start typing.
Local LLMs require setup. You need the right machine, the right model, the right tools, and enough patience to make it useful.
But beyond setup, the trade-off is more important.
Cloud LLMs
Cloud LLMs are powerful because they give you immediate access to the latest models. They are usually faster, smarter, better connected to third-party tools, and easier to scale.You benefit from huge infrastructure and constant improvement. The model gets better without you doing anything. New features appear. Tooling improves. APIs mature. Integrations grow.
That is the upside.
The downside is ownership.
Your work happens inside someone else’s system. Your workflows depend on their pricing, their policies, their model behaviour, their product roadmap, their memory limits, their data rules, and their willingness to keep supporting your use case.
You are renting intelligence.
And sometimes that is perfectly fine. Renting is convenient. Renting is fast. Renting lets you start immediately.
But you should know that you are renting.
Local LLMs
A local LLM is different.It may not be the smartest model available. It may be slower. It may require technical setup. It may not have the same tool ecosystem. It will only be as scalable as your machine, unless you invest more in infrastructure.
But it gives you something cloud models do not give you by default: ownership.
Your data stays with you. Your files can be available to the model. Your notes, research, documents, product thinking, templates, decision logs, customer interviews, strategy docs, and personal working patterns can all become part of the environment.
Over time, your local setup can become less like a chatbot and more like an extension of how you work.
Not just a clever model.
A system shaped around your judgement.
And this is where the future becomes interesting.
The 100x PM question
If AI optimists are right, we are heading into a world of agents.Not just chat windows. Agents.
AI systems that can take a task, break it down, use tools, check sources, create outputs, remember preferences, and improve over time. PMs are already overloaded, so the appeal is obvious: research agent, strategy agent, competitor monitoring agent, PRD agent, interview synthesis agent, roadmap critique agent.
More output. Less manual work. More leverage.
But where does that leverage live?
Let’s take market research.
With a cloud LLM, you start by creating the prompt. You explain your company, your market, your competitors, your product, your preferred output, the level of detail you need, what sources to trust, what to ignore, how to structure the answer, and what good looks like.
Then you refine the prompt. Then you add context. Then maybe you create a context file. Then next month, when you want to repeat the same work, you may need to rebuild a lot of it again.
Some cloud tools now have memory and project context, and that will improve. But the core relationship is still the same: you bring context to the model, the model produces the work, and the platform captures a lot of the value.
With a local setup, you still need to do the hard work upfront.
You need to define the task. You need to provide context. You need to structure the workflow. You need to explain your preferences. You need to improve the output.
But once you do that, the knowledge can stay with you.
Your research agent can know how you evaluate competitors. Your strategy agent can know what kind of market signals you trust. Your writing agent can know how you think. Your product agent can know your standards for a good opportunity, a weak assumption, or a dangerous metric.
In time, your agents can become a reflection of your professional judgement.
Not perfectly. Not magically. Not without effort.
But meaningfully.
And while they work, you can work on something else. Or think. Or talk to customers. Or do the part of product management that still requires a human brain and some taste.
Or, you know, chill.
The uncomfortable trade-off
Cloud LLMs are improving incredibly fast. They will get better memory. Better personalisation. Better tools. Better enterprise controls. Better integrations. For many PMs and many companies, they will remain the obvious choice.But every PM should understand the trade-off.
When you use a cloud LLM, you may become more productive today. But you may also be training yourself to depend on someone else’s system.
When you build local agents, you may move slower at first. But you may also be building an asset that compounds around your own way of working.
That is the real question.
Are you increasing your own professional value?
Or are you increasing the value of the cloud LLM you use?
Because in the agentic future, your salary may depend less on what you personally know, and more on what your agents can do.
And then the question becomes very simple:
Do you own those agents?
Or do you merely rent them?