What I Wish I'd Known Before I Started

If you're reading the prologue to this series and you're still at Level 0 — not using AI for coding yet, or just getting started — this post is for you. The prologue gives you the map. This gives you the packing list.

Everything here is stuff I learned the expensive way: by wasting time, wasting money, or building habits that became walls later. None of it is complicated. All of it matters.

Your Data Is the Product (Until You Pay to Change That)

Every AI provider has different policies on whether your conversations and code train their models. Some make the opt-out easy. Some bury it. Some don't offer it at all.

If you're on a free tier, be realistic — nothing is free. Your data is likely part of the deal, and that's the trade you're making for zero-cost access. If you're a paying customer, you should have clear, accessible control over whether your data trains their models. Companies that hide these controls or make them deliberately hard to find shouldn't get your money. Vote with your wallet.

That said — none of the consumer tiers we're discussing in this series are adequate for genuinely sensitive work. If you're dealing with HIPAA, PCI, classified data, or anything where a compliance officer would ask questions, you need enterprise-grade agreements with proper data residency, not a $20/month subscription. Know where the line is.

Don't Try to Outsmart the Pricing

You're not going to build a clever caching layer or an auto-routing proxy that saves you meaningful money at the solo-dev scale. I burned two weeks on this before the obvious answer won. The providers have teams of pricing engineers who've already thought of whatever you're planning. Just pick a tier that fits your usage, pay for it, and spend your engineering energy on the work that actually ships.

Pay what you can afford and nothing more. You'll know when it's time to upgrade because you'll keep smacking into the rate limits consistently — and you've ruled out that the reason is poor token hygiene, not genuine throughput demand. Let your workflow tell you when it's time, not your ambition.

Stay Away from the API (For Now)

If you're a solo dev without a business justification: the API is not for you yet. Pay-per-token pricing will punish every inefficiency in your workflow mercilessly. A sloppy prompt that wastes tokens on a subscription costs you nothing extra. That same prompt on the API costs real money, and it compounds fast.

If you absolutely must use the API, set a hard monthly spend cap on day one. Not a soft cap. Not 'I'll keep an eye on it.' A hard cap. You've been warned.

Treat the Subscription Like It Costs Per Token

Use the subscription to cut your teeth. It's low-stress and forgiving — you won't go bankrupt because a debugging loop burned through tokens. But treat it as if you're paying per token. Build tight prompts. Manage your context. Stop the agent when it's wandering. Develop the discipline now, while the cost of mistakes is capped.

This matters more than it sounds. Poor token hygiene at the subscription tier becomes a scaling wall later. The habits you build here are the ones you'll carry into Levels 3, 4, and beyond — and they either compound in your favor or against you.

Don't Marry One Provider

If you can afford it, don't limit yourself to one provider. Models are not one-size-fits-all — they have different strengths, different failure modes, and different personalities. At the time of writing (March 2026), Claude Pro and ChatGPT Plus are each $20/month, and GitHub Copilot Pro is $10/month. Running two subscriptions lets you learn which models work best for which tasks in your specific workflow. That intuition is worth more than the cost of the second subscription.

But the real value is versatility, not comparison shopping. If a provider drops the ball — ships a bad model update, degrades the service, gets weird about privacy — you can move your workflow to a competitor without starting from zero. You've already built the muscle memory. Vendor lock-in is a real risk in this space, and the best hedge against it is never letting one provider become your only option.

This versatility has a ceiling. At scale, multi-provider orchestration gets painful without API-level access, and at some point you'll probably pick a primary driver. But that's a decision you make from a position of experience, not one you fall into because you only ever used one model. In the meantime, multi-provider access also buys you outage resilience — providers go down, and when one does, you switch to another instead of staring at a loading screen.

The Short Version

Read the privacy controls before you paste code. Pay what you can afford. Avoid the API until you have a reason. Build good habits while the cost of bad ones is capped. Use more than one provider if you can. And spend your energy on the work, not the tooling economics.

---

Ready? Start with the prologue: The Agentic Maturity Model — Where Are You, and Where Are You Going?