What Is This Note?
This is a dated note on a guide we published in 2025 called "The Complete AI Copilot Stack Guide for 10x Developer Productivity". Its four-layer way of organizing AI developer tools still works as a checklist. Many of its specifics did not hold up: install commands that do not work, a product that has been renamed, model names that are out of date, and results we have no records for.
Which Claims From the Original Post Are Withdrawn?
| In the 2025 guide | What we can show | Status |
|---|---|---|
| A "$127,000 problem" of repetitive work, opened with a story about a named developer, told as real | No record supports the story or the figure | Withdrawn |
| Results from "47 teams", including 67% faster development cycles | No records support the count or the results | Withdrawn |
| ROI of 4,307%, 2,975% and 1,253% in different sections | The figures contradict each other and have no data behind them | Withdrawn |
| "10x developer productivity" | No measurement supports it | Withdrawn |
| An install command for a package named @anthropic/claude-code | That package does not exist on npm; the package is @anthropic-ai/claude-code | Corrected |
| claude-code init and claude-code setup-review commands | These commands do not exist; the installed command is claude | Withdrawn |
| A CodeWhisperer command-line setup | CodeWhisperer is now part of Amazon Q Developer | Out of date |
| Recommended models such as GPT-4 and Claude 3 | Both have newer generations; model choices change often | Dated |
The AWS announcement is on the AWS News Blog. The Claude Code package name is from the npm registry, checked on 2 October 2026.
What Still Holds Up?
The four-layer model still holds up as a way to organize the question of which AI tools to use. It is a checklist, not a benchmark.
The four layers, kept as a checklist
From the 2025 guide, without its results
Research
ideation and analysis
Design
prototyping and UI
Code
implementation and review
Deploy
release and monitoring
For each layer, ask which tool you use, how its output is checked, and how work moves to the next layer. The checking is where most of the value is.
What Do We Use Today?
As of October 2026, Claude Code does most of our implementation work, OpenAI Codex acts as an independent reviewer on larger changes, and Qodo reviews the pull requests we open. Every change passes automated gates before it can merge, and a person merges it.
The 2025 guide versus how we work in 2026
Our own setup, October 2026
- A tool for every layer, installed by command
- Productivity claimed as a multiple
- Little on how output is checked
- A small set of tools with defined roles
- No productivity multiple claimed
- Automated gates and human review on every change
This note, in numbers
As of 2 October 2026
claims withdrawn, corrected or dated
layers kept as a checklist
AI tools with defined roles in our workflow
nonexistent commands removed
Our Thread-Based Engineering post describes the gates and checkpoints in full. Our Claude Code technical debt guide covers practical patterns for keeping AI-assisted code production-grade.
The AI Copilot Stack: Questions Readers Ask
Common questions about this topic, answered.
What was the AI copilot stack guide?
A long guide we published in 2025 that organized AI developer tools into four layers: research and ideation, design and prototyping, code, and deployment. The four layers still work as a checklist. Many of the specific commands, product names and results in it did not hold up, so this note replaces it.
How do you install Claude Code?
The npm package is @anthropic-ai/claude-code, and the command it installs is claude. The guide named a package, @anthropic/claude-code, that does not exist on npm, and two claude-code subcommands that do not exist. Use the official Claude Code documentation for current install steps.
What happened to Amazon CodeWhisperer?
It became part of Amazon Q Developer. AWS made Amazon Q Developer generally available on 30 April 2024, and the CodeWhisperer product page now redirects to it. The guide described a CodeWhisperer command-line setup, which is out of date.
Did teams get 10x productivity from the stack?
We cannot show that they did. The guide reported results from "47 teams", a "$127,000 problem" and three different ROI figures that contradict each other. We have no records behind any of them, so all are withdrawn.
Which AI developer tools does Pixelmojo use today?
As of October 2026, Claude Code does most of our implementation work, OpenAI Codex acts as an independent reviewer on larger changes, and Qodo reviews the pull requests we open. Every change also passes automated gates, and a person merges it. Our Thread-Based Engineering post describes the setup.
Is the four-layer model still useful?
As a checklist, yes. Ask which tools help with research, which with design, which with code and which with deployment, and how work moves between them. The model is a way of organizing the question, not evidence that any tool improves productivity by a set amount.
Why not just update the guide?
Because AI developer tools change faster than a long tool-by-tool guide can be kept accurate. A guide full of install commands goes stale within months. A short dated note with the corrections, plus a post on how we actually work, is more useful and easier to keep honest.
What changed in this post on 2 October 2026?
We cut it down to a dated note. The results, ROI figures and the opening story are withdrawn; the wrong package name and commands are corrected; the CodeWhisperer setup is marked out of date; and the model names it recommended are flagged as dated. The four-layer model is kept as a checklist.
