Humanmaxxing
Menu

The Framework

The Maxxing Framework

Most people's AI usage plateaus in the same place: impressive on demo day, disappointing on an ordinary Tuesday. The instinct is to blame the model — or to swing the other way and hand it the keys. Both miss the actual gap, which is method. This page is the framework: what it optimizes, the principles it stands on, and the three parts that make it work.

In one sentence

A standardized, repeatable way to work with AI in software development — and to keep getting better at it.

The principles — what the framework believes

  1. Human input is the scarce term in the equation.
    AI capacity is effectively unlimited and getting cheaper; human insight is finite and expensive. Optimize for the scarce resource — never spend it where the abundant one suffices.
  2. Human value is not uniform across the work.
    It concentrates at specific points — direction, decisions, standards — and is near-zero across the toil between them. Most productivity loss is human input spent at low-value points.
  3. AI output quality is a function of your input and your reviewers.
    The equation multiplies; it doesn't add. Sharper direction raises everything downstream, and sloppy input gets amplified too. But perfect input still doesn't prevent AI mistakes — reviewers tuned to your application narrow the window.
  4. Trust is engineered, not assumed.
    Scaled output is only valuable if you can own it, and ownership requires structural verification — not confidence.
  5. The interaction compounds or decays — it never stays still.
    Every human input either evaporates at session end or becomes a permanent asset. The framework exists to make it the latter.

The three parts

Part 1

Focus

Put human judgment where it earns the most
  • Define the points in the development loop where human insight and decision-making give the most ROI — then gate those points, so work stops for the human there, and only there.
  • Treat human attention as the budget: spend it on decisions, never on toil a system can absorb.
  • Measure both clocks — human time and total time — and manage the ratio, not either number alone.

In practice: the development loops with their named human gates · two-clock convention · operating model's decision inventory.

Part 2

Craft

Continuously sharpen your craft of working with AI
  • Sharpen your craft of working with AI based on how you actually work with it — the rules, skills, and specialists come from your own sessions and corrections, not from a template.
  • Capture what works so it runs the same way every time: name your processes, scope your specialists, write the conventions down where the AI reads them.
  • Every correction becomes a permanent asset, and drift gets audited on a calendar, not on failure.
  • Grow the human too: study your own usage, sharpen your phrasing and standards — you should be a better engineer every month, not a rustier one.

In practice: the behavioral contract and layered conventions · agent roster · skills · improvement loop · maintenance calendar (<15 human minutes per touchpoint) · mining your own sessions.

Part 3

Trust

Maximize your trust in the AI output
  • Earn the trust structurally, so you can own what ships. No output self-certifies: verification is independent, blocking, and its disagreements route to human judgment — which is exactly where judgment belongs.
  • Make the checks that matter block, not advise — a check everyone assumes is a gate is an incident waiting to happen.
  • Verify against reality — run it, test it at real boundaries — never against the author's description of it.

In practice: the trust architecture · AI-checks-AI review teams · enforcement layer of blocking hooks and gates.

Measured by correct output per token — the human minutes and the machine minutes both count, and the amber never goes to zero. The numbers live in Analytics and Receipts.

Which part is your weakest?

You're busy all day but can't name your decisions — human input spent on toil → Focus is weakest.

You repeat the same corrections, session after session, and your setup works the same as it did three months ago → Craft.

You've shipped something the AI was confidently wrong about → Trust.

Use it

Self-serve

Download getting-started.claude.md and the artifacts — the framework as working files. Free, documented, yours to adapt.

Talks & workshops

The framework in 45 minutes, or a working session applying it to your team's actual workflow — receipts included.

Advisory

Standing up the three parts inside your org: the gates, the working patterns and roster, the verification structure, and the rhythm that keeps them alive.

Where this comes from: a production billing platform run by one developer at 700+ pull requests a month — every guideline on this page exists because its absence caused a real incident, documented in Field Notes. The numbers behind the claims live in Receipts.