🧭 THIS WEEK AT AI SECOND ACT
Howdy, I’ve been deep in the realms of Claude Code last few months and WOW, it is amazing. See what I’ve been up to and building below and how fast I’ve done it compared traditional SW development.
Anyone can learn it! I have a bit of an edge with a SW background, but truly anyone can do this. If you have an idea for an app, a website or whatever, you can build it!
Could be awesome at work - typically all enterprise tools suck - solution, just make one yourself!
Let me know if you’re interested in this subject area as could definitely write more about it. Next week we’ll circle back to more of an intro.
If you’re enjoying this newsletter, please share it! Forward this to a friend/colleague and have them subscribe at aisecondact.com/subscribe. Thanks!
👉 You can also just hit "Reply" and let me know what you want more (or less) of, or use the poll to give feedback.
My goal is to make this as valuable and practical as possible as we navigate the new AI era. 🚀
🧰 AI NEWS + LEARNING
Here are a few things I found recently:
Anthropic introduced healthcare focus technology into Claude. The system sucks so let’s hope this helps.
and ChatGPT did the same
Anthropic also introduced Claude Cowork - if it’s as good as Claude Code, can’t wait.
Longer tweet [or whatever that is] on being a product manager in the world of AI.
🗺️ FEATURED INSIGHT
Building at AI Speed: 600K Lines of Code in 90 Days
The Numbers Don't Lie
Between October and December 2025, I shipped five production-ready applications using Claude Code as my primary development partner. Here's what that looked like:
707 commits across all projects
603,049 lines of code written
5 complete products from zero to production
~2.4 commits per day maintaining velocity across holidays
24.7 million tokens consumed
The Portfolio
VibeBuildLab (136 KLOC) - AI-powered MVP factory. A complete monorepo platform for rapidly building and deploying AI-driven products. TypeScript/React with comprehensive tooling.
KeyFlash (29 KLOC) - Affordable keyword research SaaS. Delivers keyword data in under 3 seconds at 10x lower cost than Ahrefs or SEMrush. Built and monetized in weeks, not months.
Postrail (51 KLOC) - Social media automation for newsletter creators. Auto-generates and schedules posts across LinkedIn, Facebook, and Twitter/X from newsletter content.
Project Starter Guide (15 KLOC) - Comprehensive framework for choosing the right tech stack and architecture for any project type. Includes decision trees and best practices.
QA Architect (40 KLOC) - Framework-agnostic quality automation bootstrap. Sets up linting, formatting, pre-commit hooks, and GitHub Actions for JavaScript/TypeScript and Python projects.
What This Actually Means
Traditional estimates for these projects would have been:
Development time: 12-18 months with a team vs me: 3 mths very part time
Cost: ~ $2m - $5m in developer salaries
Actual cost: ~$650 in Claude Code fees, 3 months part time
Productivity Factor: up to 75x!!!
Cost Factor 300 - 500x!!!!
Top 10 Lessons for Claude Code Mastery
1. Break Conversations at Natural Boundaries
Don't run 500-message marathons. Create a PR, start fresh. Each conversation should focus on one feature or fix. Use /compact before switching topics.
2. Use Autonomous Agents Religiously
Stop doing manual review loops. Let /pr-review-toolkit:review-pr handle quality iteration. Use the Task tool with specialized agents (Explore, Plan, etc.) instead of doing everything yourself.
3. Always Read Before Editing
Claude Code MUST see the file before modifying it. No exceptions. This prevents hallucinated changes and ensures context-aware edits.
4. Leverage TodoWrite for Everything Complex
Breaking tasks into trackable steps isn't optional—it's how you avoid forgetting critical requirements. Use it proactively, not reactively.
5. Choose the Right Model for the Job
Sonnet 4.5: Default for 90% of tasks (5x cheaper than Opus)
Opus 4.5: Architecture decisions, complex refactoring only
Haiku: Quick reads, simple edits
Switching to Sonnet as default quintupled my monthly capacity at the same cost.
6. Symbolic Editing > File Editing
Use find_symbol, replace_symbol_body, insert_after_symbol for precise changes. Reserve replace_content regex editing for multi-line tweaks within larger symbols.
7. Never Disable Quality Gates
No eslint-disable. No --no-verify commits. No any types. Fix the root cause or don't ship it. Quality shortcuts create compound interest debt.
8. Batch Deployments, Not Every Change
Make changes locally, test thoroughly, then commit + deploy together. Don't trigger CI/CD for typo fixes.
9. Plan Mode for Complexity, Direct Action for Clarity
Use EnterPlanMode when requirements are ambiguous or multiple approaches exist. Skip it for well-defined tasks—just execute.
10. Treat AI Development Like Real Development
This isn't prototyping. It's production code. The same standards apply: testing, documentation, security, performance. Just because it's fast doesn't mean it should be sloppy.
The Real Insight
The game isn't about coding faster. It's about shipping faster. Claude Code handles the implementation minutiae—syntax, boilerplate, configuration—while you focus on architecture, user experience, and business value.
We've crossed a threshold where the limiting factor is no longer technical execution. It's decision-making, prioritization, and market validation.
The question isn't "Can I build this?"
It's "Should I build this?"
And with 90-day runway to five production apps, you can answer that question with real products instead of spreadsheets.
Built with Claude Code, Cursor, and an unreasonable belief that software development timelines are now measured in days, not sprints.
Weekly AI strategies for operating executives
— Brett
👉 Hit “Reply” and share your experience — I read every one!