Tech stack posts get a bad reputation because most of them are either sponsored or superficial. A list of logos with a sentence each doesn’t tell you anything about how someone actually works.
This is the version I’d want to read: what I use, what for, what I pay, and what I’d skip if I were starting over. I run four businesses, spend most of my working day in AI tools, and have developed clear opinions about what’s worth it and what isn’t.
No affiliate links. No sponsored mentions. Just the stack.
Why the Stack Matters More Than Any Individual Tool
Reviewing AI tools one by one misses the point. The real leverage comes from how they connect. A writing tool that feeds into a publishing workflow that connects to a distribution system is far more valuable than the same writing tool used in isolation.
The entrepreneurs I see getting the most from AI aren’t necessarily using the newest or most expensive tools. They’ve thought carefully about how their tools pass work to each other, and they’ve removed the steps that require manual handoffs. That’s the compounding part. A single tool gives you a productivity bump. A connected stack gives you a multiplier.
With that framing in mind, here’s mine.
AI for Writing and Content
Claude (Anthropic). This is my primary writing assistant across everything. I use it for first drafts, editing passes, repurposing long-form content into shorter formats, and structuring arguments before I write them. The writing quality at the top of the model range is meaningfully better than the alternatives for longer pieces, and the context window means I can paste in a full article draft and ask for a structural critique without it forgetting the beginning by the end.
For my content engine, Claude handles most of the mechanical work: turning an outline into a first draft, rewriting a blog post as an email, generating five headline options from a topic brief. I do the thinking, the point of view, and the final edit. Claude does the production.
ChatGPT (OpenAI). I still have a subscription and use it occasionally for tasks where its browsing and tool-use feel snappier. But for pure writing quality, it’s no longer my default. I find the prose it generates has a distinctive flatness that’s hard to edit out. It works better for structured outputs, tables, and data extraction than for prose.
Perplexity. For research-driven content, Perplexity is genuinely useful as a starting point. It surfaces sources quickly, gives you enough to know whether a topic is worth going deeper on, and saves an hour of tab-opening for most background research tasks. I use it before writing anything that requires knowing the current state of a topic.
AI for Research and Analysis
Research is where AI has saved me the most time and where the quality improvement is least dependent on having a distinctive voice. AI is genuinely good at summarising what exists. The judgment about what matters and what to do with it is still mine.
Perplexity Pro. My default for web research. Before any call, new business decision, or competitor analysis, I run a Perplexity query first. It produces structured summaries with citations in seconds. The alternative used to be half an hour of searching and skimming. I haven’t fully replaced that process, but I’ve compressed it significantly.
Claude with long context. For anything that requires reading a long document, contract, report, or transcript and extracting what matters, Claude’s context window is the tool I reach for. Paste in a 60-page PDF, ask three specific questions, get answers. This is one of the unglamorous but high-value use cases that doesn’t get written about much.
NotebookLM (Google). I’ve started using this for source-heavy research projects where I need to query across multiple documents at once. Less useful for quick lookups, very useful for synthesising a body of material into something I can reason about.
AI for Coding and Development
I’m not a professional developer, but I work with code regularly across my businesses. AI has changed what’s possible for people at my level significantly.
Claude Code. This is my primary coding environment now. The short version: it’s not just an autocomplete. It’s a programmable collaborator that understands your entire codebase, remembers preferences across sessions, and can execute multi-step tasks without you holding its hand through each one. I use it for WordPress plugin work, automation scripts, and anything involving more than a few files.
The customisation layer is what makes it genuinely powerful. Persistent instructions in CLAUDE.md, memory that builds up over time, reusable workflows for common tasks. Once you’ve set it up properly, it stops feeling like a tool and starts feeling like a colleague who knows your codebase.
Cursor. I still use Cursor for certain tasks, particularly when I want to stay inside a familiar editor environment with inline suggestions. It’s better than GitHub Copilot was for me. But for anything requiring deeper reasoning across files, I go to Claude Code.
GitHub Copilot. Dropped this once Claude Code became capable enough. The inline tab-completion is useful but not worth paying for separately when you’re already paying for Claude.
AI for Communication
This is the category where I’ve been more selective, because the cost of AI making communication sound wrong is higher than the cost of spending an extra few minutes writing something yourself.
Claude for email drafting. I use Claude for any email that requires careful framing. Difficult feedback, commercial negotiations, responses to complaints, anything where tone matters. I write a rough version or a brief, paste it in with context about the situation and the person, and ask for a draft. I always rewrite it in my own voice, but having a structurally sound version to react to is faster than writing from scratch.
Meeting prep. Before significant calls, I run a Perplexity query on the person and company I’m meeting, then ask Claude to summarise what I should know and suggest three questions I haven’t thought of. Takes five minutes. Reliably surfaces something useful.
I’ve deliberately not connected AI to my email in a way that sends anything autonomously. The review step matters. There are too many edge cases in communication where “close enough” isn’t good enough.
AI for Business Operations
This is where AgentVania came from. I started building AI agents for my own businesses, saw the patterns repeat across different contexts, and turned the infrastructure into something others could use. The operational side of running multiple businesses is where agents pay off most clearly.
Custom AI agents. Across my businesses I have agents handling content triage for WP RSS Aggregator, first-pass support responses, scheduled reporting, and research compilation. None of these are off-the-shelf. They’re built around specific workflows that I’d otherwise have to do manually or hire for. If you want to understand what agents actually are and how they differ from chatbots, I wrote about it at /what-is-an-ai-agent/.
Make (formerly Integromat). For connecting tools that don’t talk to each other natively. I use Make to pipe data between my CMS, email platform, and AI tools. It’s not an AI tool itself, but it’s part of the stack because AI tools without integrations are isolated.
n8n. For more complex automation workflows that need more control than Make allows. I run this self-hosted. It’s more technical to set up but gives you flexibility that SaaS tools don’t.
The honest version of AI agents for business owners is that they require upfront work to build and maintain. They’re not plug-and-play. But the payoff for the right workflows is real and it compounds. A properly built agent runs indefinitely without degrading.
What I’ve Tried and Dropped
This section is where I’d want someone to be honest with me, so I’ll be honest here.
Jasper. Used it briefly in 2023 before the major model providers caught up on writing quality. The output was template-driven in ways that showed, and the pricing didn’t make sense once Claude and GPT-4 improved. Dropped it.
AI meeting transcription tools (Otter, Fireflies, etc.). I tried several. My issue was never the transcription quality, which is fine. It was that I didn’t build the habit of reviewing the summaries, so they accumulated without being used. I switched to taking my own notes in calls and using AI to expand them afterwards. That workflow actually sticks.
AI image generation for content. I tested Midjourney and DALL-E for blog imagery. The results are technically good but I found them visually distinctive in a way that dates quickly. I still use Unsplash for photography and only use AI-generated images when they’re the right aesthetic choice for a specific piece, not as a default.
Several “AI writing assistants” built on top of GPT. Most of these are API wrappers with a pricing markup and a few prompt templates on top. Once you understand how to write a prompt, there’s very little these tools offer that you can’t do directly. I stopped paying for the layer.
The Total Monthly Cost
Being transparent here because vague claims about AI being cheap while paying for twelve tools isn’t helpful.
- Claude Pro (Anthropic): $20/month
- Claude Code (Anthropic): $100/month (Max plan, heavy usage)
- ChatGPT Plus (OpenAI): $20/month
- Perplexity Pro: $20/month
- Cursor Pro: $20/month
- Make (Integromat): ~$30/month depending on operations
- n8n (self-hosted): server costs, roughly $15/month
- NotebookLM: free
Total: roughly $225/month, or about $2,700/year.
That sounds like a lot until you compare it to what the same output would cost in human time or contractor fees. The content production alone across my sites would require a part-time hire. The coding assistance has replaced an estimated 10-15 hours of contractor work per month. The economics are clear at this level of usage.
If you’re a solopreneur just starting, you don’t need all of this. See the next section.
What I’d Recommend If You’re Just Starting
One tool, used well, beats five tools used poorly. The mistake I made early was adding tools before I’d extracted the full value from the ones I had. Resist that.
Start with a single Claude or ChatGPT subscription at the $20/month tier. Pick one high-friction task in your work and use AI to handle it for 30 days. Just one. Content drafting, email responses, research, whatever costs you the most time. Get good at prompting for that specific task before expanding.
Once that’s running smoothly and the habit is built, add Perplexity for research. Then think about automation. The automation layer requires you to understand what you’re automating first. If you haven’t done the task manually with AI assistance, you can’t build a good automated version of it.
The dead zones I wrote about separately are real: there are parts of every workflow where AI adds friction rather than removing it, and if you’re not paying attention you’ll end up with a more complicated process than when you started. Simplify before you automate.
If you want to go further with agents specifically, that’s what AgentVania does. We build the infrastructure, you get the result without the six months of trial and error.
But the most important thing is to start somewhere specific, not everywhere at once.
Related
Leave a Reply