I’m watching founders blow money on AI subscriptions they don’t need. ChatGPT Plus, Claude Pro, Midjourney, Notion AI, Jasper, Copy.ai, probably three other tools I’m forgetting. That’s $200+ per month before you’ve built anything.

Most of this is unnecessary. Some of it’s actively counterproductive because it creates dependency on tools you could easily replace with free alternatives or manual work.

Let me save you some money.

What You Actually Need

If you’re pre-revenue or early-revenue, you need exactly two paid AI tools. Maybe three if you’re doing content marketing seriously.

ChatGPT Plus or Claude Pro. Pick one. Don’t pay for both. They’re 90% overlapping for most use cases. Claude’s better at long-form writing and analysis. ChatGPT’s better at code and has more integrations. Both are $20/month. Get one, skip the other.

I use Claude Pro because I write a lot and Claude handles nuance better. But plenty of founders prefer ChatGPT because the code interpreter is genuinely useful for data analysis. Either choice is fine. Paying for both is waste.

GitHub Copilot if you’re building software. $10/month. Writes maybe 30% of your code. That percentage sounds low but it’s the repetitive boring stuff, which means you spend your brain power on the hard problems. ROI is obvious if you’re technical.

That’s it. Everything else is optional.

The Optional Third Tool

If you’re doing content marketing seriously, meaning you’re publishing multiple times per week and content is core to your growth strategy, then consider a proper writing AI tool.

But not Jasper. Not Copy.ai. Those are overpriced and honestly not better than ChatGPT/Claude for most use cases.

What’s actually useful is a tool that understands your brand voice and can generate on-brand content at scale. Newer tools like Writer or Anyword do this better than the older generation. But even these are luxury purchases. You can achieve 80% of the value with ChatGPT Plus and good prompts.

I’m not paying for specialized writing tools. ChatGPT does what I need. If I was running a content team producing 50+ pieces per month, I might reconsider. At my scale, it’s not worth it.

What You Definitely Don’t Need

Notion AI. Notion’s free tier is fine for most small teams. The AI features are incremental convenience, not game-changers. If you’re considering paying for Notion AI, just use ChatGPT in a separate tab. Copy/paste is not that hard.

Midjourney. It’s $10/month minimum and honestly for most startup use cases, free tools like Ideogram or Bing Image Creator are good enough. If you need custom imagery regularly, Midjourney’s worth it. But most founders I talk to paid for a month, generated 15 images, and then never used it again.

Grammar/writing tools beyond basic Grammarly. Grammarly’s free tier catches typos. That’s 90% of the value. The premium features are nice-to-haves. Same with tools like ProWritingAid or other writing assistants. ChatGPT will edit your prose for free.

Specialized AI tools for [specific function]. There’s an AI tool for everything now. AI for meeting notes. AI for email. AI for scheduling. AI for CRM. AI for recruiting. Most of these are just ChatGPT with a wrapper and a $29/month subscription.

Before paying for any specialized AI tool, ask yourself: can I do this with ChatGPT and five minutes of manual work? Usually yes.

The Free Tier Strategy

Here’s what I’m using for free:

Claude.ai free tier for quick questions and drafting. The free tier gives you enough messages per day for casual use. I only hit the limit if I’m doing heavy work, in which case I switch to my Pro account.

ChatGPT free tier as backup when I hit Claude’s limits or when I need image generation or web search. It’s slower and you don’t get GPT-4 consistently, but it works fine for simple tasks.

Bing Chat for web search with citations. It’s actually better than ChatGPT’s search function and completely free.

Google’s Gemini for when I need to analyze really long documents. Gemini has a massive context window on the free tier. I use it specifically for reading long PDFs or analyzing contracts.

GitHub’s free Copilot tier if you’re working on open source projects or if you’re a student. Not applicable to most founders but worth knowing about.

This combination covers probably 80% of what I use AI for. I only drop into paid tools when I need higher quality output, faster responses, or more usage volume.

When to Upgrade

You should upgrade from free to paid when you’re hitting limits that cost you time or quality.

If you’re using ChatGPT free and constantly getting “you’re using GPT-4 too much, here’s 3.5 instead,” and that’s slowing you down, upgrade to Plus. $20/month is less than the productivity hit.

If you’re doing client work and the quality difference between free and paid Claude actually impacts deliverables, upgrade to Pro.

If you’re building a product and Copilot would save you five hours per week, that’s $50-100 of your time depending how you value it. Spending $10 for that is obviously worth it.

But don’t upgrade preemptively. Use free tiers until they hurt. Then upgrade exactly the tool that’s causing pain.

The Team Calculation

If you’ve got a team, this math changes. You might need multiple seats. That gets expensive fast.

Be selective about who gets paid accounts. Your technical cofounder needs Copilot. Your content person needs ChatGPT Plus or Claude Pro. Your designer might need Midjourney.

But your intern probably doesn’t need paid tools. Your part-time contractor probably doesn’t either. They can use free tiers or share accounts (check terms of service, some tools allow this, some don’t).

One approach I’ve seen work: get two paid ChatGPT Plus accounts for the team to share. Not technically allowed per OpenAI terms, but realistically they don’t enforce it for small teams. Safer approach: get ChatGPT Team tier at $30/user/month if you’ve got 3+ people who need it regularly. Cheaper than individual Plus subscriptions once you’re past two users.

API Access vs. Subscriptions

Different calculation for API access. If you’re building AI features into your product, you’re paying per token rather than subscribing.

This can get expensive fast or stay cheap depending on your architecture. I’ve seen founders blow through $1000/month in API costs for products with 50 users because they didn’t optimize their prompts. I’ve also seen founders serve thousands of users for $100/month because they cached aggressively and kept prompts tight.

If you’re using APIs:

• Start with Claude or GPT-3.5. Don’t jump to GPT-4 or Claude Opus until you’ve proven the cheaper models can’t do what you need.
• Monitor costs obsessively. Set up billing alerts. Check usage daily.
• Optimize prompts for length. Every unnecessary word costs money.
• Cache responses when possible. Don’t re-generate the same content repeatedly.
• Consider fine-tuning if you’re doing the same task thousands of times. Upfront cost but cheaper per-use.

The Build vs. Buy Question

Here’s the uncomfortable truth: most AI features you’re considering building, someone else has already built better and cheaper.

If you’re thinking “I’ll build an AI assistant for [specific task],” stop. Google “[task] AI tool” and you’ll find five SaaS products doing exactly that. Use one of those instead of building your own.

Build AI features only when they’re genuinely differentiated and core to your value prop. Otherwise just integrate existing tools.

I’ve watched too many founders spend six months building AI features that they could’ve integrated via API in a week. The custom build isn’t better. It just costs more and takes longer.

What I’d Do Starting Today

If I was starting a new company tomorrow with zero existing AI spend, here’s what I’d do.

Month 1-3: Use entirely free tools. ChatGPT free, Claude free, Gemini, Bing Chat. See which one I use most. Prove I’m actually getting value before spending money.

Month 4: Upgrade to ChatGPT Plus or Claude Pro (whichever I was using more). Keep everything else free.

Month 6: Add GitHub Copilot if I’m writing code regularly. Keep everything else free.

Month 12: Reassess. Am I hitting limits? Do I need team accounts? Is there a specialized tool that would save significant time?

This approach forces you to prove value before spending. Most founders do the opposite. They subscribe to everything day one, use 10% of it, and keep paying because canceling feels like admitting waste.

The Real Cost

The financial cost of AI tools is actually not the big problem. Even if you’ve got $200/month in AI subscriptions, that’s not going to kill your runway.

The real cost is distraction. Every new tool is something to learn, to integrate, to maintain. Every subscription is another login to remember, another billing alert to ignore, another email from their marketing team.

Your time and focus are more valuable than the subscription cost. Don’t adopt tools unless they clearly save more time than they consume.

Final Take

Most founders are overthinking AI tools. The technology is powerful but you don’t need every tool that launches. You need one good LLM (ChatGPT Plus or Claude Pro), maybe Copilot if you code, and discipline to keep the free tier of everything else.

That’s $20-30 per month. Everything beyond that should have a clear ROI that you can articulate in one sentence. “This tool saves me X hours per week doing Y task, which is worth $Z to me.”

If you can’t articulate that, you don’t need the tool. Unsubscribe and put that money toward something that actually moves the business forward.