Claude Is My Architect. Gemini Is My Contractor. | Back Office Contrarian

Here's something the AI productivity gurus won't tell you: using Claude for everything is a fast way to burn through your session limit, hit a wall at 10 AM, and either sit there waiting or pull out your credit card to keep working.

I figured out a better way. And it didn't come from a YouTube video or a guru's paid course. It came from sitting in my workflow, watching the usage meter climb, and asking the question every back office thinker eventually asks — is there a smarter way to build this?

There is. And it involves treating two AI tools the way a smart contractor treats a project: one person designs the blueprint, one person swings the hammer. You don't pay an architect to swing the hammer a hundred times. You pay them to get the blueprint right once.

The Problem With Using Claude For Everything

Claude is extraordinary at reasoning. It pushes back, refines, catches what you missed, and improves every pass through a draft. That back-and-forth iteration is exactly what makes it powerful — and exactly what makes it expensive in terms of token usage.

Anthropic runs Claude on a session-based usage system. You get a five-hour rolling window, and when you hit the ceiling, you either wait for it to reset or pay to top off and keep going. As of late March 2026, that burn rate got faster — Anthropic confirmed that during peak weekday hours (8 AM to 2 PM Eastern), sessions drain more quickly than before due to surging demand on their infrastructure. About 7% of users are hitting limits they never hit before.

If you're doing heavy, repetitive production work inside Claude — generating the same type of content over and over from a finished prompt — you're burning your most expensive resource on your most predictable task. That's the opposite of leverage.

The fix is simple once you see it. Stop using Claude as a factory floor. Use it as a design studio.

The Architect/Contractor Split

Here's the workflow I use to build every article on this blog and every SEO piece for the pattern business I co-own with my business partner Tracey at Sunrays Creations:

Claude is the architect. I bring Claude a raw idea. We go back and forth — sometimes for a long session, sometimes across multiple sessions. I push back on what it gives me. It pushes back on what I give it. The prompt gets sharper. The structure gets tighter. The voice gets more aligned with what I'm actually trying to say. We iterate until the master prompt is bulletproof.

Gemini is the contractor. Once that master prompt exists, it goes into Gemini. I feed it the topic, the angle, the affiliate search terms — the variables that change from piece to piece. Gemini executes. It produces publish-ready content from the spec Claude helped me build. Same quality output. Zero Claude tokens burned on a task that doesn't require Claude's reasoning anymore.

The prompt is the framework. Build it once with the right tool. Execute it repeatedly with the efficient one.

If that philosophy sounds familiar, it should. It's the same thing I wrote about in The Framework Doesn't Change. Only The Tools Do. Covey didn't update his framework every time a new app launched. The framework was right. The tools that serve it are interchangeable.

What This Actually Looks Like In Practice

The Back Office Contrarian article creation prompt took real iteration to build. It specifies voice, structure, AdSense placement rules, Amazon affiliate box format, FAQ requirements for AEO optimization, schema markup, author bio placement, label taxonomy — the whole stack. That level of specificity didn't come from one session. It came from building, testing, noticing what was off, and refining.

Claude was the right tool for that work. That's high-judgment, high-iteration work that benefits from a reasoning partner who pushes back.

But once that prompt existed? Every article brief I feed into Gemini comes out close to publish-ready. The heavy thinking already happened. Gemini is executing a tested spec, not inventing one from scratch.

The same system runs at Sunrays Creations. We sell cross stitch patterns on BigCommerce. The SEO content strategy for the store needed a master prompt tailored to that niche — product-specific language, pattern-related search terms, the right affiliate hooks. Claude helped build and refine that prompt. Gemini runs it every time we need a new piece of content for the store.

One prompt. Two businesses. Infinite executions. Zero extra Claude tokens.

The Money Angle Nobody Is Talking About

This isn't just a productivity hack. It's a cost management strategy.

Claude Pro runs $20 a month. Max 5x is $100. Max 20x is $200. If you're burning your session on repetitive production tasks during peak hours, you're paying premium prices for commodity work. Worse, you might be topping off — paying API overage rates on top of your subscription — to finish work that Gemini would handle for free.

Gemini has a generous free tier. For content production from a finished prompt, it performs remarkably close to what Claude produces from the same input. Not identical — Claude's reasoning in the design phase is genuinely superior. But in the execution phase, from a locked-in master prompt? The gap closes significantly.

The math is straightforward. Design with the premium tool. Produce with the free one. Protect your Claude sessions for the work that actually requires Claude.

Screenshot, Don't URL-Drop

One more token conservation habit worth building: stop pasting URLs into Claude and asking it to go read a page.

When you drop a URL, Claude fetches and processes the entire page — headers, footers, nav menus, sidebar content, ads, the whole thing. That's a heavy token burn for what is usually a small slice of useful information. A screenshot of the relevant section gives Claude exactly what it needs and nothing it doesn't. You'll be surprised how much faster your sessions move when you make this one switch.

Same principle applies to long documents. Paste only the section that's relevant to your current question. Don't upload the whole file when you need Claude to look at two paragraphs. Every token you don't burn on setup is a token available for actual work.

The Books That Shaped How I Think About This

The architect/contractor mental model didn't come from an AI course. It came from decades of thinking about how work gets organized and who should be doing what.

David Allen's Getting Things Done is the foundational text here. The core principle — capture everything, process deliberately, execute from a trusted system — maps directly onto the prompt workflow. The master prompt IS the trusted system. You build it once. You trust it. You execute from it without having to re-think the structure every time.

James Clear's Atomic Habits is the second piece. The habit loop — cue, craving, response, reward — is exactly what a production workflow runs on. The master prompt is the cue. The content output is the reward. When the system is reliable, the habit becomes automatic. You stop thinking about the process and start producing at volume.

And if you want to go deeper on the AI side of the workflow, there's a growing shelf of prompt engineering books worth your time. The ones worth reading focus on systems — how to build prompts that hold up at scale — not collections of one-off hacks that stop working when the model updates.

Why The Gurus Don't Tell You This

The AI productivity space is full of people selling the dream of using one tool for everything. Pick your platform, go all in, and the content practically writes itself. That's the pitch.

What they don't tell you is that going all-in on a single AI tool for both high-judgment design work and repetitive production work is the most expensive way to operate. You're using a scalpel to dig a ditch.

The contrarian move — the back office move — is to match the tool to the task. Use the reasoning-heavy, premium tool for reasoning-heavy work. Use the capable, free tool for execution once the thinking is done. Protect your session budget for the conversations that actually change the output.

Most small business owners won't do this because it requires thinking about their AI workflow as a system rather than a convenience. But that's exactly the difference between using AI and leveraging AI.

Build It Once. Run It Forever.

Here's the part that makes this feel like passive income for your workflow: a good master prompt doesn't expire.

The article creation prompt I use for Back Office Contrarian took real time and real iteration to perfect. I could not have built it without Claude's back-and-forth. But now that it exists, it produces consistent, on-brand, monetized content every single time I run it. The investment was in the design phase. The returns compound in the execution phase.

That's the architect/contractor model in its final form. The architect earns their fee once, on the blueprint. The contractor executes from it indefinitely. You want to be the person who owns the blueprint — not the person re-inventing the structure every time they need a room built.

If you're not already building your workflows this way, start with one prompt. Pick your highest-volume, most repetitive content task. Take it to Claude. Iterate until it's right. Then take that finished prompt to Gemini and run it there going forward.

Your Claude session budget will thank you. Your production volume will go up. And you'll have built something that keeps working long after the session that created it has closed.

That's the back office move. The stuff the gurus forgot to mention — probably because they want you to keep buying their AI tool subscriptions.