⚙️ THE WORKFLOW
I've been tracking every AI tool I've tried since early 2023. Current count: 47 tools evaluated, 6 still in active use. The others were cancelled after a trial period or replaced by something better. Here's what made the cut and why, plus the evaluation workflow I use before paying for anything new.
My AI tool evaluation workflow:
Node | What it does |
|---|---|
Trial signup | Free tier or 14-day trial, never pay upfront |
Week 1 test | Use it for 3 real tasks in my actual workflow |
Time log | Measure actual time spent with vs without the tool |
Integration check | Does it have an API or n8n node? Can it fit into automations? |
Decision gate | Is it saving ≥30 min/week? Keep. Less? Cancel. |
Baserow log | Record tool, use case, verdict, monthly cost |
This sounds mechanical. It is, deliberately. AI tools are easy to get excited about and easy to keep paying for out of FOMO. Measuring actual time saved cuts through that.
The 6 tools that made it:
1. Claude (Anthropic) — strategy and writing
Everything that requires sustained reasoning: client proposals, complex debugging, writing longer-form content. The difference between Claude and competitors is quality of reasoning on ambiguous problems, not speed. I pay for Pro ($20/month). It pays for itself on a single good client proposal per month.
2. Google Gemini Flash (via AI Studio API) — workflow AI
Every AI node in my n8n workflows runs Gemini Flash. It's fast, cheap (fractions of a cent per call), and good enough for classification, summarisation, and structured JSON output. I spend under $5/month on API calls across all my automation workflows. This is the tool I'd recommend to anyone building AI-powered automations on a budget.
3. Cursor — coding
The AI-native code editor that replaced VS Code for me in early 2024. The tab completion is genuinely better than GitHub Copilot for the workflows I write. The codebase-aware chat (not just file-aware) is the feature that matters. $20/month.
4. ElevenLabs — voice
I use this exactly once per month: to narrate the script for my YouTube workflow walkthroughs. One voice, cloned from my own recordings, consistent across every video. $5/month on the Starter plan. The quality gap between ElevenLabs and every other TTS tool is still significant.
5. Perplexity — research
When I need current information (competitor pricing, recent product updates, regulatory changes), Perplexity is faster than Google and cites its sources. I've replaced about 60% of my Google searches for research tasks. $20/month for Pro, which I justify by the time saved on client research.
6. Blotato — social scheduling
Every piece of content I create gets repurposed to 4–6 platforms via n8n + Blotato. I write once, the workflow handles the rest. $49/month. The ROI here is less about time saved and more about consistency — I was previously posting sporadically because manual posting felt like a chore.
Total monthly AI spend: $114. Revenue attributable to tools above: measurable on Claude and Cursor, fuzzy on the rest. The honest answer is I've stopped trying to calculate ROI on every tool and instead set a hard budget: $150/month max for AI tools. Anything new has to displace something existing.
🔧 THE STACK MOVE
Google AI Studio — the cheapest AI API for automation workflows
Most AI workflow tutorials default to OpenAI. OpenAI GPT-4o costs approximately $2.50 per million input tokens and $10 per million output tokens. Gemini Flash costs $0.075 per million input tokens and $0.30 per million output tokens — roughly 30× cheaper for comparable quality on structured tasks.
Price: Free tier includes 15 RPM and 1,500 requests/day — enough to run a small automation stack without paying anything. Paid tier is pay-as-you-go with no monthly minimum.
The honest tradeoff: Gemini Flash is excellent for classification, JSON generation, and summarisation — the most common automation AI tasks. It's noticeably weaker than GPT-4o or Claude Sonnet on tasks requiring complex reasoning, ambiguity handling, or creative output. For automations that need to make nuanced judgements (like complex lead qualification or long-form content generation), the cheaper model will disappoint you. The right approach: use Gemini Flash for the 80% of workflow AI tasks that are structured, and use a stronger model via API for the 20% that require real reasoning.
📡 THE SIGNAL
Gemini 2.5 Pro is the model to watch in 2025 — blog.google
The jump from 1.5 to 2.0 to 2.5 has been faster than expected. Gemini 2.5 Pro's performance on coding and reasoning benchmarks is now competitive with GPT-4o and Claude Sonnet 3.5 at a fraction of the API cost. The automation use case is particularly well-served — structured output via JSON mode has gotten more reliable with each release.
The AI subscription fatigue is real — latent.space
A pattern I've noticed in conversations with other indie operators: they're subscribed to 8–10 AI tools, actively using 2–3, and can't tell you what the others do. The tool layer is getting crowded fast. The operators who get value from AI are the ones who've built deliberate evaluation criteria, not the ones chasing every new release.
Claude's extended context is changing how I work — anthropic.com/blog
Being able to paste an entire codebase or a stack of client documents into context and ask questions across all of it is genuinely new. It's not just "longer context" — it's a different mode of working with information. The use case I hadn't anticipated: pasting my entire n8n workflow JSON and asking Claude to explain what it does, find potential failure points, and suggest improvements.
Your next 100 customers are already in Apollo
Find, reach, and close your perfect deals — without juggling five tools or hiring more reps.
Apollo gives you everything you need to build real pipeline, fast. From inbound to outbound, first touch to close.
All in Apollo.
![Theo [Founder]](https://media.beehiiv.com/cdn-cgi/image/format=auto,width=256,fit=scale-down,onerror=redirect/uploads/user/profile_picture/05538803-f858-4c87-aeb1-e1c6ca95e8e2/Photo_on_13-3-25_at_12.33__2.jpeg)
