shared/cover-letter/README.md

Cover Letter

Two versions of the same message.

• short-wake-up-letter.md — 5-minute read. The blunt version. Read this if you have 5 minutes.
• field-guide-manifesto.md — 30-minute read. The full case. Read this if you need to convince yourself or someone else.

Both make the same argument: - You should have started last year. - Every quarter of waiting compounds against you. - Starting is not hard. Starting is deciding.

If you're already convinced, skip both and start Level 1.

shared/cover-letter/field-guide-manifesto.md

The Field Guide

A longer letter about what you're missing, what it's costing you, and what to do about it.

The setup

You've heard the announcements. Every quarter there's another model. Every month there's a new capability. Every week someone on LinkedIn claims 10x productivity.

You've tried ChatGPT. Maybe Claude. Maybe Perplexity. You got a nice email once. A meeting summary. A code snippet. And you moved on.

That's the trap.

The people and companies pulling ahead did not stop at "I tried it." They built systems around it. They wrote personal system prompts. They procured enterprise accounts. They ran evals on outputs. They shipped agents. They measured cost per resolved item. They trained their teams.

And they compounded.

What compounding looks like

Year 1: You save 30 minutes a day. That's 130 hours in a year. Year 2: Your prompts are 3x better. You save 90 minutes a day. That's 390 hours. Year 3: You have agents. Your team has agents. Some workflows are 80% automated. You save 3-4 hours a day. That's 900-1200 hours.

Meanwhile, the person who "tried it once" is still googling the same questions, writing the same emails, taking the same meeting notes.

The gap is not linear. It is exponential.

The real-world problems you are already facing

If any of these describe you, you're already paying the cost of waiting.

For individuals

• Inbox is winning. You spend 90+ minutes a day on email you don't remember writing. An AI triage + draft workflow reduces that to 20 minutes.
• You are the bottleneck. Every project waits on your review or your draft. AI turns you into a reviewer of first drafts, not the writer.
• You lose context between meetings. You can't remember what you decided last week. An AI weekly review + notes system fixes it in 15 minutes a week.
• You research the same thing twice. No prompt library, no knowledge base — every task starts from zero.
• You can't keep up with your field. Papers, releases, competitor moves — you feel behind. An AI research routine catches you up in 30 minutes a week.
• Your best work is inconsistent. Some emails are sharp, some are muddled. A personal system prompt and voice guide fixes that permanently.
• You are afraid to use AI at work. No AUP, no data classification — you don't know what's safe. So you don't use it. So you fall further behind.

For companies

• Your top performers are already using AI on the side. Some are 3x more productive quietly. Some are leaving for companies that let them use it openly.
• Your CX team is drowning. 40% of tickets could be triaged or drafted by AI. Every ticket you don't automate is a hire you shouldn't have to make.
• Your sales team is inconsistent. New reps take 6 months to ramp. A sales copilot with your playbook and battle cards cuts that in half.
• Your legal team is a bottleneck. Every contract review is manual. Clause extraction and NDA screening reduce turnaround from days to hours.
• Your finance team runs on tribal knowledge. Month-end close, variance analysis, reconciliation — all repeatable, all automatable, all still manual.
• Your engineering team is losing velocity. AI-assisted development is now table stakes. Teams without Copilot or Cursor ship measurably slower.
• Your marketing content is generic. Every AI-shy team writes copy that reads like it was drafted in 2019. Every AI-fluent team ships 5x more, on brand, with SEO built in.
• Your customer data is not being learned from. Support tickets, sales calls, product feedback — all a goldmine, all sitting in silos, all unlearned.
• Your board is asking about AI strategy. You have a slide. You do not have a program.
• Your customers are asking about AI features. You have a roadmap item. You do not have a shipping plan.
• Your competitors are shipping. Every week another player in your space adds an AI feature. Every quarter another RFP asks about your AI capabilities. Every year the market rewards the ones who moved.

If three or more of these describe your company, you are Level 0 and you are behind.

The three groups in the market right now

Group A — Frontier operators (5%). Have been building since 2022. Multi-agent systems in production. Custom evals. Fine-tuned models. They win outsized market share.

Group B — Active adopters (25%). Started in 2023-2024. Have governance, copilots, and shipping workflows. They're keeping pace and closing on Group A.

Group C — Still figuring it out (70%). Bought some licenses. Sent people to conferences. No policy, no evals, no shipping. They are losing ground quarterly, and most don't know it yet.

The gap between Group B and Group C doubled in the last twelve months. The gap between Group A and Group B is widening. If you're in Group C, the window to close the gap by trying harder alone is closing. You now need a program, not effort.

Why now, specifically

Three things changed in the last 18 months that make waiting far more expensive than it used to be:

1. Cost per useful output dropped 10x. What cost $1 to produce in 2024 costs $0.10 now. If you weren't using AI because "it's expensive," that reason is gone.
2. Reliability crossed a threshold. Frontier models now handle complex multi-step reasoning, tool use, and long context reliably enough for production. What used to require prompt engineering hero moves now works out of the box.
3. Ecosystem matured. MCP, agent frameworks, eval tools, observability platforms — the plumbing is now real. You don't have to build everything. You just have to install the pieces and configure them.

The excuses that were valid in 2023 — cost, reliability, tooling — are not valid in 2026.

The counter-argument, honestly considered

"But hallucinations." Real. And solvable. Grounded generation (RAG), verified sources (Perplexity), and human-in-the-loop review reduce hallucination cost to near zero for production workflows. If you're still worried about hallucinations, you're using AI wrong.

"But security." Real. And solvable. Enterprise contracts with zero-retention, data classification matrices, and tool tiering handle 95% of the risk. The other 5% is process, not technology.

"But my team isn't ready." They're readier than you think. The average person can be Level 1 in 30 days. The blocker is not talent. It is a program.

"But we tried it." You didn't. You gave people a login and hoped. That's not a program. That's a wish. This kit is the program.

"But we're too small." Small teams benefit most. An AI-fluent 10-person team runs like a 30-person team. The economics are more favorable at your scale than at 10,000 people.

"But we're too big." Big companies benefit hardest. Every process you have is documented. Every process that's documented can be systematized. Every process that can be systematized can be AI-augmented.

"But we don't have AI engineers." You don't need them for Levels 1-2. Level 3 needs one platform engineer. Levels 4-5 need a small applied research team. Most companies never need to go past Level 3.

The prescription

Read the AI Fluency Program in this order:

1. This letter — you're done.
2. /README.md — 5 minutes. The map.
3. /LEVEL-MAP.md — 10 minutes. Where you're going.
4. /level-1-essentials/ — 30 days. The foundation.
5. /level-2-intermediate/ — 60-90 days. You become a builder.
6. Everything above Level 2 is optional depending on your goals.

For companies, add:

• /level-1-essentials/company/01-governance/ on Day 1 (before any employee uses AI)
• /level-1-essentials/company/10-rollout-playbook/ for the 30-60-90 plan
• Steering committee on Day 1
• Two department pilots by Day 60

What Level 1 actually feels like

• Day 1: You draft an AUP. You classify your data. You procure ChatGPT and Claude.
• Day 7: You have a master system prompt. You've written 3 personal prompts. You save 30 min a day already.
• Day 14: You've routed 20 tasks to the right models. You've caught 2 hallucinations. You know the jargon.
• Day 21: Your team has a copilot. Two workflows are in beta.
• Day 30: You have wins to show. You know what's next.

None of this is theoretical. This is what the ~30% of adopters have already done. It is not hard. It just requires starting.

What Level 2 buys you

• Custom agents on your own stack, doing real work
• A RAG system over your own docs answering questions your team keeps asking
• An MCP server exposing your data to any AI client
• Cost per workflow measured in dollars, not vibes
• An eval set that catches regressions before your customers do

What Level 3+ buys you

• AI products shipping to real customers
• Multi-agent workflows replacing full-time work
• Category-defining features
• Recruiting advantages
• Moat

The final honest note

This program does not require you to be technical. Levels 1 and much of Level 2 are approachable by anyone willing to work through them.

It does require you to start. Reading is not starting. Watching a webinar is not starting. Buying a license is not starting.

Starting is: opening level-1-essentials/00-start-here/ and completing Day 1 today.

The models will keep getting better. The cost will keep dropping. The tools will keep multiplying. None of that helps you if you never begin.

You should have started last year. Start today anyway.

Written to be useful in 2026 and every year after. If you're reading this in a future year, replace "last year" with "the year before whenever you actually read this" — the message doesn't change.

shared/cover-letter/short-wake-up-letter.md

The Short Version

You should have started last year. Start today anyway.

If you're reading this in 2026, here's the honest picture:

Your competitors are already three cycles ahead. Not because they're smarter. Because they started earlier and compounded. Every quarter you don't build AI fluency, they close another deal you didn't know was in play, ship a feature you're still scoping, and answer a customer you're still triaging.

The tools are no longer the moat. The habits are. Anyone can buy ChatGPT Enterprise. Not everyone can run a weekly review with it, route the right task to the right model, evaluate outputs against a real eval set, or ship an agent that actually saves 20 hours a week.

Waiting is not neutral. It is a decision to fall behind at 20-30% a year, compounding.

The five things you lose by waiting

1. Time. The average knowledge worker with a functional AI setup saves 5-10 hours a week. That's 250-500 hours a year. Every year you wait, you leave that on the table.
2. Judgment. The instinct for "what's a good AI task, what's a bad AI task, what's a hallucination risk" only develops through use. You cannot cram it before a deadline.
3. Institutional knowledge. Your codified prompts, department copilots, and RAG systems compound. A company that started in 2024 has 2 years of eval sets, prompt versions, and observed failure modes. You cannot buy that.
4. Leverage on your best people. Your top performers are already using AI on the side. Without a program, they either become 3x more productive quietly (and get poached) or they leave for a company that lets them use it openly.
5. Bargaining power with vendors. Companies with active AI programs get better contracts, roadmap input, and early access. Late arrivals pay list and take defaults.

What "started last year" actually looks like

• A written AI acceptable-use policy signed by everyone
• A 4-tier data classification anyone can recite
• ChatGPT Enterprise, Claude, and Perplexity procured with DPAs
• Every department has a copilot with an eval set
• Two agents in production with observed cost per resolved item
• A steering committee that meets monthly
• Every new hire gets 5 hours of AI training in their first week
• The company has retired 3 workflows because AI made them obsolete

If you have none of these, you are Level 0. Ninety days from now, you can be Level 1 across the board with real work behind you. But only if you start.

The one action

Read field-guide-manifesto.md for the full picture, or open /level-1-essentials/00-start-here/30-day-starter-sequence.md and begin Day 1.

The kit is done. The path is drawn. The only variable left is you.

This is a cover letter for a five-level AI fluency program. If you're already using AI daily, skip to Level 2. If you don't know what a system prompt is, start Level 1. Everyone else, read the field guide.