CONTEXT WINDOW
SANITATION & SANITY

Right now, software teams are stuck in the hangover of the "tokenmaxxing" era—caught between AI hype that promised instant 10x velocity and a reality where senior engineers spend their days debugging sprawling, hallucinated slop.

Watching teams burn millions of tokens only to grind their velocity to a halt has made a lot of good engineers skeptical of agentic coding altogether. Set against that frustration, I have a potentially contrarian take: the hype is actually right—you practically never have to write code by hand again. And mourning the end of manual syntax typing makes about as much sense as mourning the end of hand-written Assembly when C compilers arrived.

The catch is that once you stop typing code yourself, your job changes overnight: you are now an involuntary high school teacher, and underneath the mystique, coding agents behave practically like teenagers:

• Oracular Wisdom: They will always give you a confident "trust me, bro" answer—even when you have no idea how they arrived at it or whether they'd give you a completely different answer five minutes from now.
• Short Attention Span: Even when your directions are crystal clear, they might decide you meant to put the cutlery in the microwave and go eat some Tide pods. Dump 50 files into their context window and they immediately lose the plot.
• Never Cleans Their Room: Just like the 2021 Nature Lego study, LLMs systematically overlook subtractive changes—leaving a sprawling mess of wrappers and duplicate code behind unless you check the rubric yourself.

AGENTIC LORD OF THE FLIES

Spend five minutes online and you'll hear a seductive promise: "Don't worry about architecture or clean code anymore—the agents will figure out how to organize things."

Having taught both high school English and computer science, I can tell you what happens when you drop a group of bright, energetic teenagers onto an island with zero adult supervision and assume they'll self-organize:

"Drop a swarm of unsupervised teenagers onto a codebase and you don't get a self-organizing parliamentary democracy. You get Lord of the Flies."

For the first twenty minutes, the agents hold up the conch shell and write a very polite PLAN.md. By week two—because LLMs wake up with stateless amnesia in every new session, copy whatever broken windows are already in their context window, and suffer from relentless addition bias—the signal fire is out, Piggy's glasses are smashed (except Exception: pass), three rival tribes have built competing auth_helper_v2.py modules because there was no CODEBASE_MAP.md, and your repository is a 4,000-line monolith.

You don't fix an island full of teenagers by yelling harder at the prompt or throwing more tokens at the beach; you put an adult in the room and run a structured classroom: Design Your Course (clean exemplars, descriptive feedback, and self-assessment tools), Prepare Your Materials (an isolated sandbox and curated context window), and Check the Rubric (end-to-end verification and continuous Socratic red-teaming).

GREAT HUMAN CODE ≠ A GREAT AGENT CLASSROOM

What happens when you run generate_report_card.py against some of the most respected open-source Python, TypeScript, and JavaScript repositories in the world?

Frameworks like FastAPI, Flask, and Anthropic's official Model Context Protocol TypeScript SDK (mcp-ts-sdk) are world-class human codebases—and they score 100/100 (A+) on traditional human-era dimensions like type annotations, test assertion density, and naming clarity.

Where they get taxed is Agent Ergonomics. A human developer in VS Code can Cmd+Click to line 4,200 of fastapi/routing.py (6,553 lines) in zero milliseconds. An AI coding agent has to ingest ~65,000 tokens of context window just to read that single file—and without a CODEBASE_MAP.md or SCHEMA_DICTIONARY.md, it burns dozens of tool calls grepping the repo blind before it even gets there:

In the agentic era, a unified monorepo is table stakes—an LLM cannot reason across a dozen scattered repositories without hallucinating cross-service contracts. And you shouldn't have to risk production uptime while your team learns a new way of working.

Instead of trying to reform Jack's tribe while production is on fire, we never touch prod during the training. Over a 4-part, month-long sprint (four weekly half-day working sessions), your engineering team leaves Jack's camp behind and builds Ralph, Piggy, and Simon's camp from the ground up: standing up a brand-new, clean monorepo wired to an isolated agentic dev environment where the shelters actually stand, the signal fire stays lit, and you fall back in love with your code. At the end of the month, you walk away with a superior, production-ready parallel application—tested, codified, and agentically managed—that your team can cut over to for a "soft landing" completely at your own leisure.

TRAINING & ENGAGEMENT FORMATS

• The Codebase Report Card (Offline Audit & Written Roadmap): Want an independent second opinion before bringing the whole team into a room? Grant read-only repository access (and optionally your recent cloud & AI billing breakdown) for an asynchronous deep-dive. You receive a candid, jargon-free written assessment identifying where your architecture is bleeding cloud dollars and AI context tokens, your highest-risk "meat puppet" zones (monolithic files, AST clones, undocumented contracts, un-specced mocks), and a prioritized 30-day action plan—followed by a 45-minute walkthrough call (with the audit fee credited directly if you upgrade to a PD Day or Monorepo Makeover).
• The 1-Day PD Day (Standalone Socratic Red-Teaming Clinic): Professional Development for involuntary teachers—requiring zero infrastructure changes. In a single hands-on day with your existing repository and AI tools, your team installs a baseline AGENTS.md syllabus and AST context packer (pack_context.py), learns how to replace sycophantic leading prompts with open-ended Socratic Red-Teaming, and ships immediate subtractive cleanups live in the room (creditable directly toward the 4-Part Monorepo Makeover).
• The 4-Part Monorepo Makeover (Flagship): Four weekly half-day live working sessions + asynchronous PR and harness review between sessions. Without ever touching your live production environment, your team stands up a parallel, fully tested, SSM-isolated monorepo in an agentic dev sandbox—ready to cut over to on your own schedule.