Building Your AI Stack: What to Use, What to Pay For, and When to Stop
AI tools are multiplying fast. There are platforms for writing, platforms for coding, platforms for images, platforms for research, platforms for specific workflows, platforms built into tools you already use. The number is absurd and growing. A new “must-have” tool drops every other week. Newsletters declare winners. Benchmarks contradict each other.
Most people who feel overwhelmed by this aren’t missing information. They’re missing a framework. And the most common mistake isn’t picking the wrong tool. It’s treating the tool question and the money question as separate decisions when they’re actually the same one.
This article handles both together, because that’s the only way to handle either one well.
By the end, you’ll have a clear answer to: what AI should I actually be using, and what should I be paying for it?
What You’ll Learn Here
- How to audit what you’re already paying for before spending anything new
- A simple two-category mental model for every AI tool on the market
- How to pick a generalist platform and the right tier for your actual use
- A decision tree that handles the most common scenarios in plain yes/no questions
- When specialists make sense and the math that should drive that decision
- How to evaluate any AI tool or subscription, including quarterly checkpoints
- Who the high-tier plans are actually for (and who they’re not for)
- How usage-based API costs work and how they can compound unexpectedly
- When to switch tools and when to stay put
- Specific, opinionated recommendations by user type
Section 1: Check What You Already Have
Before you evaluate a single new tool or consider spending a dollar, spend fifteen minutes on this question: what AI am I already paying for?
The uncomfortable reality is that a meaningful percentage of people reading this are already paying for capable AI and have never turned it on.
Here’s what to check:
Google Workspace (any paid plan): You have Gemini. It’s integrated into Gmail, Docs, Sheets, Drive, and Meet. Workspace Business Standard includes Gemini features. If you use Google for work, this is already there.
Microsoft 365: You have Copilot built into Word, Excel, PowerPoint, Outlook, and Teams. The availability varies by plan, but if you’re on a business Microsoft 365 subscription, AI assistance is part of what you’re paying for.
Notion Business: Notion AI is included. It can write, summarize, and edit directly inside your workspace.
Canva Pro: AI design tools are included. Magic Studio handles background removal, image generation, and design suggestions.
GitHub: Copilot is available to all GitHub users and free for students and open-source maintainers. If you write code at all, check your GitHub account status before paying for a separate coding AI.
The action here is simple: spend one week using whatever AI you already have access to before doing anything else. Discover its actual limits through actual use before deciding you need something different. You might find it handles most of what you need. You might find clear gaps. Either way, you’re making the next decision with real information instead of assumptions.
Section 2: The Two-Category Framework
Every AI tool in existence fits one of two categories.
Generalists are chat platforms that handle a wide range of tasks in plain language. ChatGPT, Claude, and Gemini are the big three. You describe what you need, they do it. Writing, analysis, coding, brainstorming, research, explanation. They do a lot of things well enough to be genuinely useful across most situations.
Specialists are tools built for a specific job. An AI that only makes slides. An AI that only handles legal documents. An AI built into your code editor. An AI for image generation. An AI that processes receipts. Specialists go deeper in one area and trade breadth for precision.
The mistake most people make is trying to find the right specialist before they’ve built a solid foundation with a generalist. Specialists are optimization tools. They amplify a capable base. They don’t substitute for one.
This framework matters because it changes how you think about every tool you encounter. When you see a new AI product, the first question is: is this a generalist or a specialist? If it’s a specialist, the second question is: do I have a specific, frequent limitation that this actually solves?
Get the foundation right first. The specialists come later, if at all.
Section 3: Picking Your Generalist
This is where the tool decision and the cost decision become the same decision. There is no “which platform should I use?” question that isn’t also a “what should I pay?” question. They’re answered together.
The Platform Choice
The three major options are ChatGPT (OpenAI), Claude (Anthropic), and Gemini (Google). In March 2026, they are genuinely close in capability. Obsessing over benchmarks is the wrong approach. The smarter question is: which one fits where you already work?
Start with Gemini if you live in Google Workspace. Gmail, Docs, Drive, Sheets are your home. Gemini’s integration is genuinely deep in ways no other platform can match. Using it is often a matter of clicking the Gemini icon inside a tool you’re already in.
Start with Claude if your work is writing-heavy, you work with long and complex documents, or you want something that’s thoughtful and careful in its responses. Claude tends to handle sustained work on complicated tasks better than the alternatives.
Start with ChatGPT if you want broad all-around capability, the most mature ecosystem of third-party integrations, and the widest range of built-in tools. It’s the safe default for general use.
Don’t overthink the platform choice. You’re not locked in. Pick one and spend real time with it. The platform that fits your existing workflow is almost always the right one.
Start on the Free Tier
Always. No exceptions to this rule.
Free tiers in 2026 are not crippled demos. They’re genuinely capable. The limitations are real but they’re not artificial ones designed to frustrate you into upgrading. The free tier is what it is: a fully functional version with usage limits.
Here’s what you get on each platform’s free tier right now:
- ChatGPT Free: GPT-5.2 Instant model, limited to roughly 10 messages per 5 hours before throttling kicks in. Ad-supported as of early 2026.
- Claude Free: Sonnet 4.6 with low usage limits. You’ll hit the ceiling during heavy use.
- Gemini Free: Gemini 2.5 Flash, 100 AI credits included.
Spend two weeks on the free tier doing real tasks. Not toy problems. Actual work you’d do anyway: drafting emails, summarizing documents, thinking through decisions, writing code. When you start using AI daily and hitting rate limits, that frustration is the correct signal to upgrade. Not FOMO. Not because a newsletter said to.
The $20/Month Decision
When the free tier’s limits genuinely get in your way, the $20/month tier on any of the major platforms is one of the highest-ROI subscriptions most professionals can buy.
Here’s a simple version of the math: if AI saves you two hours per week and your time is worth $25 per hour, that’s $200 of value for $20 spent. You need to save roughly one hour per month to break even at any reasonable rate. For daily users, that’s easy.
Current pricing across the major platforms (verify at each platform’s pricing page before subscribing, as pricing changes frequently):
| Platform | Tier | Price | What You Get |
|---|---|---|---|
| ChatGPT | Plus | $20/month | GPT-5.2 Thinking, Sora access, DALL-E 4, 5x usage limits, Advanced Voice |
| Claude | Pro | $20/month ($17 annual) | Sonnet 4.6 with 5x usage vs. free, Research mode, Google Workspace integration, extended thinking, MCP connections |
| Gemini | AI Pro | $19.99/month | Gemini 3, 1,000 AI credits, full Workspace integration, 1 month free trial |
| Gemini | AI Plus | $7.99/month | Newer lower-cost tier, modest step up from free |
| ChatGPT | Go | $8/month | Unlimited GPT-5.2 Instant, no advanced features, no ads |
| Grok | SuperGrok | $30/month | Grok 4 access, higher limits within X ecosystem |
Pricing verified March 2026. See ChatGPT’s pricing page, Claude’s pricing page, and Google’s AI plans page for current details.
One note on Grok: SuperGrok costs 50% more than the comparable tiers on the main three platforms and lacks the workflow integrations of the others. There’s no strong reason to start here unless you’re already deeply embedded in the X ecosystem.
The flip side of the math: if you’re not using AI daily, or you can’t name specific tasks where it saves you real time, don’t upgrade yet. The right time to upgrade is when the limits frustrate you, not when it feels like the responsible move.
Section 4: The Decision Tree
Work through this honestly. Think about what you actually do, not what sounds useful in theory.
Q1: Are you using AI regularly right now?
- No: Start on the free tier of one generalist. Use it for two weeks on real work before considering anything else. If you haven’t found a use case in two weeks, you don’t need to pay for anything yet.
- Yes: Go to Q2.
Q2: Are you hitting the free tier’s limits?
- No: Stay on free. The limitation is likely prompting skill, not the tier. Invest time in getting better at asking before spending money on higher limits.
- Yes: Upgrade to the $20/month tier of your current platform. Don’t switch platforms. Don’t add a second one yet.
Q3: Are you using your $20/month platform daily and still running into walls?
- No: You’re in the right tier. Stay.
- Yes: Name the specific wall. Is it usage limits? Evaluate the heavier tier. Is it a capability gap for a specific task? That might be a specialist question, not a tier question. Go to Q4.
Q4: Is there a specific task you do frequently (weekly or more) that your generalist handles poorly?
- No: The limitation is probably your approach, not the tool. Invest in prompting and workflow before adding spending.
- Yes: That’s the signal to evaluate a specialist. Go to Section 5.
Section 5: Adding Specialists
The specialist decision uses the same framework you just worked through. No new logic required. It’s the same question applied to a different layer of your stack.
When Specialists Actually Make Sense
You hit a clear, specific limitation with your generalist that costs you time or quality. That limitation happens frequently, not occasionally. You can name exactly what the specialist does better. “It’s better at AI” is not an answer. “It generates slides that don’t need 30 minutes of reformatting” is an answer.
Real examples of specialist cases worth evaluating:
- Presentations weekly: If you’re making slides regularly and your generalist output always needs extensive cleanup, a dedicated AI slide tool like Gamma or Beautiful.ai might recover that time and justify the cost.
- Image generation as core output: If you’re a designer and image generation is central to your workflow, Midjourney at $10 to $30 per month has a clear ROI.
- Daily coding work: If your generalist’s code suggestions consistently miss context or need heavy editing, Cursor or GitHub Copilot at $10 to $19 per month likely pays for itself. See the Claude Code article if you want to go deeper on AI-assisted coding.
- Document processing at scale: If structured data extraction from documents is part of your job, a specialized tool built for that workflow will beat any generalist consistently.
The Math to Run Before Adding Any Specialist
- How often do I actually do this task per week?
- How much time does the specialist save me per instance?
- What does the specialist cost per month?
- Time saved per week x my hourly rate x 4 weeks = monthly value. Does that number comfortably exceed the subscription cost?
If yes by a clear margin: the specialist is defensible. If it’s close: try the free tier first. If there isn’t a free tier, the tool probably isn’t confident enough in its value to earn yours.
The Two-Subscription Cap
Most people should cap AI subscriptions at two.
Slot one: your main generalist platform. Slot two: either a second generalist you genuinely use for distinct workflows, or one specialist you use heavily.
Everything else should come from free tiers, tools you’re already paying for, or project-based one-off access. Paying for three or more subscriptions without a clearly articulated reason for each is almost always waste.
Section 6: The Evaluation Framework
Use this to evaluate any AI tool or tier at any time, not just at the start.
The Four-Question Evaluation
1. The frequency test. How often do you actually use this? Monthly use doesn’t justify a subscription. Weekly might. Daily use almost certainly does.
2. The specificity test. Can you name exactly what this does better than what you already have? Vague answers (“it feels smarter”) don’t count. Specific answers do: “It doesn’t lose context on 50-page documents,” “its code suggestions actually run without edits the first time,” “it connects to my calendar and surfaces the right things automatically.”
3. The workflow fit test. Does this tool work where you already work, or does it want you to change your process to come to it? Good AI tools fit into your existing workflow. Bad AI tools demand that you adopt theirs.
4. The cancellation test. Can you easily cancel if it stops being useful? Set a calendar reminder for 90 days from when you subscribe. Treat that reminder as a genuine evaluation point, not a formality.
The Quarterly ROI Checkpoint
Run this every three months on every paid AI subscription:
- Are you still using this daily?
- Can you name specific tasks where it saved you time this month?
- Are you paying for features you’ve never used?
- Has your workflow changed in ways that make a different tool more relevant now?
- What would you actually lose if you canceled today?
The red flag test: If you can’t remember the last time a paid AI tool did something genuinely useful for you, that’s your answer. Cancel it. You can always resubscribe.
Section 7: The $200/Month Question
High tiers exist. Some people genuinely need them. Most people considering them are doing so for the wrong reasons.
Here’s what the high tiers include as of March 2026:
| Platform | Tier | Price | What It’s For |
|---|---|---|---|
| ChatGPT | Pro | $200/month | Unlimited GPT-5.2 Pro, no usage caps, max reasoning compute, full Sora 2 Pro, expanded Codex agent |
| Claude | Max 5x | $100/month | Sonnet 4.6 with 5x more usage than Pro, Opus 4.6 access with 1M context window, agent teams preview |
| Claude | Max 20x | $200/month | All above with 20x more usage than Pro; the tier for Claude Code power users doing extended coding sessions |
| Gemini | AI Ultra | $249.99/month | Gemini 3 Pro, 25,000 AI credits, Veo 3.1, 30 TB storage, YouTube Premium bundled |
| Grok | SuperGrok Heavy | $300/month | Grok 4 Heavy, elevated limits for demanding use |
Note on the Claude Max $100/month tier: it occupies useful middle ground between the $20 Pro tier and the full $200 tier. For heavy Claude Code users or anyone consistently hitting Pro limits, this is worth evaluating before jumping to $200.
Verify current pricing at claude.com/pricing before subscribing, as these tiers change.
Who the High Tiers Are Actually For
- People using agentic tools like Claude Code or OpenAI Codex daily who consistently hit usage limits
- Developers whose primary workflow is AI-assisted coding for hours each day
- Researchers doing AI-assisted analysis at scale
- Professionals who can point to specific revenue or output that the higher tier directly enables
- Teams sharing a high-tier subscription where per-person cost makes sense
The Honest Test
If you have never hit your $20 tier’s usage cap, you do not need the $200 tier. The jump from free to $20 is transformative for most users. The jump from $20 to $200 is incremental for almost everyone. The only people who need to be at $200 are those doing something specifically heavy where usage limits actually break their workflow.
The FOMO trap is real. High tiers feel like the serious, committed choice. They’re not. They’re the right choice for a narrow set of use cases. Upgrading because it feels more professional is exactly the logic the companies are counting on.
Section 8: Usage-Based Costs and How They Sneak Up
Most readers using AI through the chat interface will never need to think about this. But if you use AI through other software or automations, read carefully.
When Usage-Based Costs Apply
Usage-based pricing comes into play when:
- You access AI through third-party tools or automations that call AI APIs behind the scenes
- You build something (even using no-code tools) that triggers AI calls automatically on a schedule or trigger
- You move beyond manual chat use into anything that runs without you initiating it
How Token Pricing Works
A token is roughly three-quarters of a word. 1,000 tokens is approximately 750 words. Current API pricing ranges from around $0.20 per million tokens for lighter models to $15 or more per million tokens for frontier models. This sounds negligible until you have automations running hundreds of times daily.
The email automation example: You build a workflow that summarizes incoming emails and saves them to a database. Each email processes roughly 2,000 tokens. You get 50 emails per day.
- Daily cost: 50 emails x 2,000 tokens x $0.02 per 1,000 tokens = $2/day
- Monthly cost: approximately $60/month
That automation, which felt trivial to build, costs three times what a subscription would. The math isn’t wrong if the value is there. But you need to know the math before you build, not after.
The key question: are you using AI through other software or automations? If yes, find out how those tools bill AI usage. Track it for a month before building more automations.
For more on building AI automations, see the No-Code Automation guide.
Subscriptions vs. API: When Each Makes Sense
- For personal, daily use: Subscriptions almost always beat pay-as-you-go
- For sporadic or low-volume automated use: API might be cheaper
- For high-volume automation: Do the math explicitly before building
Section 9: When to Switch Tools
People switch tools too often. The cost of switching isn’t the money. It’s the lost prompting habits, the saved context, the workflows you’ve built, and the learning curve you have to re-climb.
Valid Reasons to Switch
- Your current tool has a genuine capability gap for something you do frequently, and another tool actually solves it
- Your workflow has fundamentally changed and a different platform genuinely fits it better now
- The tool has gotten meaningfully worse over time, not just different
- You’ve given it an honest extended try and it truly doesn’t fit how you think
Not Valid Reasons to Switch
- You saw a benchmark ranking another model higher
- A newsletter or tech site called something “the best AI right now”
- A new model was released and the demos look impressive
- You’re curious what else is out there
“X is better” is not a reason to switch. “X can do Y, which I need daily and my current tool cannot” is a reason to switch.
Section 10: Practical Recommendations by User Type
Casual Users (a few times per week, varied tasks)
Start on a free tier of ChatGPT or Claude. Use it for two weeks on real work.
Upgrade to one $20/month subscription when you’re hitting limits or want better models. No more than that.
Warning: don’t upgrade before you’ve built a real habit. The free tier is genuinely capable for infrequent use. Pay when the limitations start costing you something.
Daily Users (AI is a regular part of work)
Start with a $20/month subscription on one platform. Choose based on workflow fit, not benchmark rankings.
Add a second subscription only if you can name specific tasks that your primary platform handles poorly. Adding a second platform out of curiosity is $240/year of shallow value.
Warning: the $200/month tier is almost certainly not for you unless you’re hitting usage caps daily.
Power Users (multiple hours per day, coding, agentic workflows)
Start at $20/month, upgrade when you hit limits consistently, not preemptively.
If you use Claude Code or OpenAI Codex heavily: the Claude Max 5x tier at $100/month is worth evaluating before jumping to $200. The specific question to ask: are you losing hours of productive work to usage limits hitting mid-session? If yes, upgrade. If no, don’t.
Power user does not mean high spender. The goal is still ROI, not spending more.
Google Workspace Users
Check your plan first. Gemini features may already be included or available as an add-on to your existing subscription.
Use the Gemini integration that’s already there. It’s genuinely embedded in Workspace in ways no standalone platform can match for people who live in Google’s tools.
Only add a second platform if there’s a specific, named task that Gemini consistently handles poorly.
Teams and Organizations
Team plans, at roughly $25 to $30 per seat per month, usually make more sense than individual subscriptions when you need shared workspace, admin controls, and data privacy guarantees.
Evaluate actual usage patterns before buying seats for everyone. A mix of team subscriptions for heavy users and free tiers for occasional users often beats a blanket rollout. Paying for 20 seats when 6 people use it daily is expensive underutilization.
For the full comparison of platforms and their team pricing, see the Platform Breakdown. For AI capabilities already built into your existing tools, see AI In Your Tools.
Section 11: The Anti-Recommendation
Don’t pay for AI if:
- You haven’t used the free tier long enough to know what you’d use it for
- You’re paying “just in case” or because it feels responsible
- You already have AI through tools you’re paying for and haven’t tried it yet
- You can’t name what you’re getting for the money
Don’t upgrade tiers if:
- You’ve never hit a usage cap on your current tier
- You’re not sure what the higher tier includes that your current one doesn’t
- You’re doing it because it feels like the serious, committed move
Don’t add subscriptions if:
- You’re using each existing one less than once a week
- You can’t articulate what each one does that the others don’t
- You’re subscribing to “stay informed” or “try different models”
The principle underneath all of this: the best ROI in AI tools comes from going deep on one platform, understanding where it’s strong and where it falls short, and building it into your actual workflow. Dabbling across five platforms is expensive shallow value. Mastering one is cheap deep value. This is not an abstract principle. It’s what separates people who get enormous value from AI from people who spend money on it and feel vaguely disappointed.
Conclusion: The Simple Version
The whole framework in seven steps:
- Check what AI you’re already paying for before doing anything else.
- Pick one generalist platform. Don’t overthink it.
- Spend two weeks on the free tier doing real work.
- Upgrade to $20/month when the limits get in your way, not before.
- Add a second platform or a specialist only when you hit a real, specific, frequent limitation.
- Evaluate quarterly. Cancel what you’re not using.
- The $200/month tiers are for a narrow slice of heavy users. You’ll know if you’re there.
Everything else in this article is the reasoning behind those seven steps.
Pricing data in this article was verified in early March 2026. AI subscription pricing changes frequently. Always verify current pricing at chatgpt.com/pricing, claude.com/pricing, and one.google.com/about/google-ai-plans/ before subscribing.