Free brief · 22-point checklist

Your prototype works. Will it survive production?

Built fast with AI coding tools? This 15-page engineering brief shows what still stands between your working prototype and real users, and how far you have to go.

Instant PDF download. No sales call.

22 checks across behavior, security, architecture, and operations
Scoring bands that tell you what your count means
Third-party research from GitHub, DORA, METR, and Veracode

By Brian Geisel, Founder & CEO
Geisel Software has built production-ready software for NASA, Teledyne FLIR, and others since 2011.

Cover of The AI Coding Reality Check, an engineering leadership briefing from Geisel Software.

// The production gap

The build is done. The decisions are not.

A working prototype can show up remarkably fast. It authenticates. It queries. It renders. It demos.

Then the questions change.

What happens when the network disappears mid-operation?

What happens when two writes hit the same record?

What happens when an input arrives that nobody tested?

What happens when traffic hits ten times what you planned for?

What happens when a user can reach data that is not theirs?

Those are not simply code-generation problems. They are engineering decisions, and they determine whether something that works in a demo can hold up in production.

// The research

Working code is not the same as production-ready code.

55%

Faster on bounded greenfield work

GitHub

90%

Of developers now use AI at work

DORA

45%

Of AI-generated code fails security tests

Veracode

24%

Substantially trust what AI writes

DORA

These are separate measurements from separate studies, each on its own population and task mix. The brief carries the full citations, sample sizes, and the context that decides what each figure does and does not support.

// What is inside

Seven places to look before you call it production-ready.

Behavior

Does it work when reality stops cooperating?

Dependencies

Do you know what the system actually depends on?

Security

Can it survive inputs designed to break it?

Maintainability

Can somebody safely change it six months from now?

Architecture

Was the architecture chosen, or did it emerge?

Operations

Can you tell when it is failing and recover it?

Testing

Will a test catch it before your users do?

AI can accelerate the work. It cannot define done.

Get the brief →

// 22-point assessment

How far is your prototype from production?

The brief includes a production-readiness assessment across all seven areas. Count the checks your build cannot pass and see where the biggest gaps are.

Git history scanned for secrets, not just the tree.
Someone other than the model can explain it.
A rollback has been performed, not documented.

Three of the twenty-two. The other nineteen, and the scoring bands that tell you what your count means, are in the brief.

Get all 22 checks →

Page 11 · Production readiness

// Inside the brief

Research, not a sales deck.

Fifteen pages of third-party research on where AI helps, where it quietly costs, and what has to be true before an AI-built system is safe to deploy.

Page 05 · The roleThe code got easier, the engineering did not. What shrinks as generation accelerates, and what grows.

Page 08 · SecurityVeracode ran 80 tasks through 150+ models. The output compiles, runs, and demos, and 45% still fails security analysis.

Before it meets real users, find the gaps.

Use the AI Coding Reality Check to evaluate your prototype across behavior, dependencies, security, maintainability, architecture, operations, and testing.

Get the brief →

15 pages · 22 production-readiness checks · Research from GitHub, METR, DORA, GitClear, Veracode, and others.

// When the assessment finds a gap

Do not start over. Finish what you built.

Knowing what is between your prototype and production is useful. Closing those gaps is the next step.

The Geisel Prompt-to-Production Sprint starts with the prototype you already have, identifies the production risks that matter, and focuses the engineering effort on getting it ready to deploy.

3–6 weeks

Fixed engagement.

$25,000

Fixed price.

1 working prototype

Hardened, tested, and moved toward a deployable baseline in your own git.

The Sprint starts with a 30-minute Viability Review of the actual application. We identify the biggest production risks, decide whether the Sprint is a fit, and tell you what we would attack first.

Explore the Prompt-to-Production Sprint → Get the brief first