My minimalist AI stack: the only tools that actually matter when you work alone

Eighteen months ago, I had seventeen tabs open permanently. Each one was an AI tool I was “testing”. One for writing, one for summarising, one for generating images, one for transcribing, one for automating, one for taking notes, one for slides, one for… you get the picture.

I was spending more time evaluating tools than actually producing anything with them.

This is the solopreneur paradox in 2025: you have access to unprecedented firepower, and that abundance itself becomes the problem. Every week, a new “15 best AI tools for freelancers” list floods LinkedIn. Every month, a new tool promises to change everything. And in the meantime, the real work waits.

Today, my AI stack fits in the palm of one hand. It’s the most productive decision I’ve ever made.

The tool-hopping trap: how testing 20 tools destroys your focus

Tool-hopping — that habit of jumping from one tool to the next without ever mastering any of them — isn’t due diligence. It’s procrastination dressed up as professional curiosity.

Here’s what actually happens when you adopt a new AI tool:

There’s an invisible entry cost. Creating an account, understanding the interface, learning which prompts work, integrating the tool into your workflow, deciding where to store outputs — all of it chips away at your cognitive capital.

The learning curve is longer than you think. Using Claude or ChatGPT at 20% of their capabilities is easy. Using them at 80% — knowing when to call on them, how to structure a prompt for your specific context, spotting their blind spots — takes weeks of regular practice. Every new tool starts from zero.

Fragmented attention kills depth. An effective solopreneur stands out by how deeply they exploit their tools, not by how many they have. Depth requires repetition and familiarity. It’s incompatible with constant change.

Context costs are real. With five tools for five tasks, you lose continuity. Your working context — projects, style, constraints — doesn’t exist coherently in any of them. You start from scratch every session.

The result: weeks spent “optimising your workflow” without shipping more. Subscriptions piling up. And a low-grade anxiety: what if you miss the next tool that changes everything.

The truth: the next tool changes nothing if you haven’t yet exhausted the potential of the ones you already have.

The 4 criteria for an AI tool to earn a place in your stack

Before adding anything, I run every tool through four questions. One negative answer, and the tool stays out.

1. Does it replace something I’m already doing, or does it create a new task?

A good AI tool should save you time on an existing task — not tempt you into doing things you weren’t doing before.

If it pushes you to create videos when you never made them, to produce infographics when that’s not your format, it expands your work surface without expanding your revenue. You drown.

2. Does it integrate into your current workflow without friction?

A tool that requires you to change how you work has a high adoption cost. It should slot in where you already work, reduce back-and-forth, and not create a new data silo.

3. Can you measure its impact within two weeks?

After two weeks of regular use, if you can’t clearly say what the tool gained you in speed or quality, it has no place in your stack.

4. Is its value-to-cost ratio (time + money) positive from the first month?

Cost isn’t just the subscription. It’s also the time spent learning and integrating it. If the tool doesn’t generate net positive value in the first month of serious use, there’s a problem.

My current stack: what I use, what I cut, and why

Here’s what I use today. Not an exhaustive list — just what’s in my stack, why it’s there, and what I eliminated.

The reasoning tool: Claude (Anthropic)

Claude is my main tool. Not for generating content at scale — for thinking.

I use it to break down complex problems, sharpen arguments, challenge my assumptions, draft first versions, structure proposals, prepare interviews.

What sets it apart: its ability to maintain long, coherent context, its tendency to nuance, and the quality of its reasoning on subtle topics.

What I cut: ChatGPT running in parallel “for comparison”. Pure tool-hopping. One LLM, mastered well, delivers more than two tools alternated.

The transcription and capture tool: Whisper / local solution

Everything I say — meetings, out-loud thinking sessions, calls — gets transcribed automatically. I no longer take notes during calls. I talk, transcribe, and have Claude process the transcript.

This workflow has changed the quality of my presence in meetings: when you’re not busy taking notes, you actually listen.

I use a local Whisper-based solution for privacy reasons. Near-zero cost, excellent reliability, direct integration.

What I cut: Otter.ai, Fireflies, and two other transcription tools. They all did the same thing with separate subscriptions.

The automation tool: Make (formerly Integromat)

Make connects my tools and automates low-value repetitive tasks: sending a follow-up email after a signature, creating a task when a client replies to a form, archiving paid invoices.

It’s not strictly an AI tool, but it amplifies the impact of the others by eliminating friction.

What I cut: Zapier (too expensive), N8N (too much maintenance), and a dozen automations built for tasks that weren’t worth the effort.

The augmented research tool: Perplexity

For quick research and monitoring, Perplexity replaced my back-and-forth between Google and an LLM. It combines real-time web search with usable synthesis.

I use it for recent factual questions, verifying information before a proposal, and quick research before a call.

I don’t use it for deep work — that’s Claude’s role.

What I cut: Three AI newsletters I’d stopped reading, an extension that summarised web pages, an RSS aggregator I spent two hours configuring and never looked at again.

What I cut and why — the honest debrief

Over eighteen months, I tested and eliminated:

Two image generation tools (Midjourney, DALL-E): I don’t produce visual content at volume. When I need it, I pay a designer or use a stock image library.
A slide generation tool: my slides stay simple. Keynote is enough.
Two AI-augmented “second brain” tools: I spent more time feeding the system than extracting value from it.
An AI assistant specialised for accounting: my accountant uses their own software. Adding an AI layer created friction.

The common thread: these tools addressed problems I didn’t have, or had already solved another way.

How to evaluate whether a new tool is really worth the adoption cost

New releases aren’t going to slow down. Here’s the protocol I apply now.

The 48-hour rule

When a tool catches my eye, I note the name and wait 48 hours. If after 48 hours I still think it’s worth it, I carry on. If not, it was passing curiosity.

This rule eliminates 80% of the tools I would have tested impulsively.

The precise use-case test

Before opening an account, I need to be able to say: “I’m going to use this tool for [specific task] that I currently do [frequency] and that takes me [time]”. If I can’t, I have no real use case.

The serious trial period

If a tool passes both filters, I give it two weeks of intentional, regular use. After two weeks, I ask: would I produce less, lower quality, or more slowly without this tool? If the answer is yes, it enters the stack.

The replacement cost

Before adopting, I ask: what does it replace in my current stack? If the answer is “nothing, it’s an addition”, that’s a red flag. A minimalist stack only grows by substitution.

Real AI productivity isn’t about the number of tools

There’s a stubborn misconception: that the most productive people have the most sophisticated stack. That mastering twenty AI tools is a competitive advantage.

It’s the opposite.

The most effective solopreneurs I know have simple stacks. They know their tools deeply. They’ve developed stable workflows that require no deliberation. Their brain stays free for strategy, client relationships, creation.

Deep mastery of an AI tool means knowing exactly how to prompt it for your specific context, having battle-tested prompts, knowing its limits. It’s a skill built over time, through repetition — and it’s incompatible with constant change.

Fewer tools, used more often, understood more deeply: that’s the formula. It’s not spectacular. It doesn’t generate viral LinkedIn content. But it produces results.

Your ideal AI stack probably isn’t the one you haven’t tried yet. It might already be there — underused, waiting for you to finally give it your full attention.

Need a hand building something — a website, a SaaS, or an AI automation? Sébastien de Bollivier, the dev behind SEK, can help.

Also worth reading: learning to code as a parent-teen duo · solopreneur & AI stats 2026.

Frequently asked questions

How many AI tools should I use at most as a solopreneur?

Keep your stack strictly to 4 AI tools. Beyond that, you dilute your focus and lose 30% of your productivity in constant switching. Simple rule: one tool per key function — writing, research, automation.

Why does a minimalist AI stack deliver better results?

You gain depth and mastery with just 3 well-chosen tools. Every additional tool adds noise without real value. Fewer tools means 2x more concrete results on your solo projects.

How do I choose the tools in my AI stack without spreading myself too thin?

Test each tool for 7 days before adding another one. Prioritise those that cover 80% of your needs with a smooth integration. Avoid any tool that doesn't solve a specific problem in your daily workflow.

Got an idea to ship? A website, a SaaS, an AI automation — built with you.

Talk about your project