2026 BLS Data: An AI Coaching Assistant Guidelines Playbook for All 306,500 U.S. Coaches and Scouts — Borrowing Stanford's CS336 CLAUDE.md

It is a Tuesday evening, mid-season. You have just wrapped a two-and-a-half-hour practice and you are sitting in the empty locker room reviewing tomorrow morning's opponent film. Your phone buzzes. Your starting point guard has screenshotted a conversation with ChatGPT: he pasted last week's turnover clip description, and the AI handed him a tidy 12-step "correct movement" breakdown with a closing line that says he should "run this in the next game." You stare at it for five seconds and realize something every coach in this country is about to face: when an AI coaching assistant starts going around you to feed athletes the answers, every one of the 306,500 coaches and scouts in the U.S. needs a written set of AI coaching assistant guidelines that turn the model into a co-pilot, not a stand-in head coach. On June 1, 2026, the Stanford CS336 "Language Modeling from Scratch" team open-sourced a 74-line CLAUDE.md file that gives the industry the cleanest AI-agent teaching-boundary template available — and it is the perfect starting point for coaches who want to bring AI training assistants onto the practice floor without giving up the chair.

According to the U.S. Bureau of Labor Statistics' August 2025 Occupational Outlook Handbook update, Coaches and Scouts (SOC 27-2022) held 306,500 jobs in May 2024, with employment projected to grow 6% from 2024 to 2034 (faster than the 3% all-occupations average) and 41,800 openings per year on average. The 2024 median annual wage is $45,920. Roughly 64% work in educational services, 22% in arts/entertainment/recreation, and 10% are self-employed. That BLS profile tells you something important: most U.S. coaches are not sitting in front-office video rooms with full analytics staffs. They are part-time, seasonal, low-budget operators — which means the temptation and the risk of unmanaged AI training assistants both hit them harder than they hit the NBA, NCAA Division I, or English Premier League.

1. Pain-Point Deep Dive: BLS Data Exposes Three Pressures on Coaches and Scouts

Data shows that three pressures are squeezing the U.S. coaching workforce in 2026, and every one of them is documented in the BLS text itself.

Pressure 1: Part-time pay plus marathon hours leaves no analyst on the bench. BLS reports the May 2024 median annual wage for coaches and scouts at $45,920 — with the lowest 10% earning less than $27,490. The Work Schedules section explicitly notes that "part-time work is common" and that full-time coaches "may work more than 40 hours a week for several months during the sports season." Research shows the 64% of coaches in educational services and 22% in arts/entertainment/recreation rarely have a dedicated analyst. The full list of duties BLS spells out — "plan practice sessions, analyze the strengths and weaknesses of individual athletes and opposing teams, keep records of athletes' and opponents' performances" — lands on the coach personally. That is precisely the surface area where an AI coaching assistant can either double a coach's leverage or quietly inflate the work into hallucination territory.

Pressure 2: Athletes are already using AI to route around coach authority. BLS lists one of the coach's core qualities in the Important Qualities section as the ability to "motivate, develop, and direct athletes to help them reach their potential," with duties spelled out as "instruct athletes on proper techniques, call plays, make decisions about strategy." In practice, 2026 youth athletes are already using ChatGPT, Claude, and Gemini to generate practice plans, post-game self-reviews, and "how do the pros train this" lookups. The moment the AI just hands back the answer — for example, telling the athlete "you missed because your inside foot was slow, switch to a drop step next time" — the athlete shortcuts the coach's correction cadence and skips the technique-internalization process BLS keeps describing. For the first time, the profession is sharing decision authority with a non-employed invisible assistant.

Pressure 3: Scouting-report workflows have no shared rules, and AI output quality is all over the map. BLS describes the scout workflow as "research news media and other sources, attend competitions, view videos of the athletes' performances, and study data about the athletes to determine their talent and potential." Research shows that more and more high school and small-AAU scouts now pour player-video transcripts, stat sheets, and game clip notes straight into an LLM and ask it to draft a full scouting report. The problem: without shared AI coaching assistant guidelines, the same raw inputs yield NBA-grade prose one day and Reddit-tier hot takes the next.

2. What the News Actually Says: Stanford CS336 Publishes an AI Agent Teaching-Boundary Template

On June 1, the Stanford CS336 "Language Modeling from Scratch" course team committed a 74-line CLAUDE.md to the public repository stanford-cs336/assignment1-basics. It is addressed explicitly to "AI coding assistants (like ChatGPT, Claude Code, GitHub Copilot, Cursor, etc.) working with students in CS336" and it nails the AI agent's role in a single line: "Primary Role: Teaching Assistant, Not Solution Generator" — the AI is to function as a teaching aid that helps students learn through explanation, guidance, and feedback rather than by completing assignments for them. Source: AI Agent Guidelines for CS336 at Stanford.

The structure is three rule blocks. What AI Agents SHOULD Do lists concept explanation, pointers to lecture/handouts/official documentation, code review with directional suggestions, guiding questions for debugging, explanation of Python/PyTorch/CUDA/Triton error messages, and sanity checks via toy examples. What AI Agents SHOULD NOT Do is a hard line: no Python or pseudocode, no problem solutions, no completing TODO sections, no editing of the student repo, no running bash, no full refactors into finished solutions, no pointers to third-party implementations. Teaching Approach lists six concrete actions: ask clarifying questions, reference concepts, suggest next steps, review their code, explain the why, prefer tests and invariants over fixes. Three Good/Bad dialog examples anchor the abstract rules to copy-pasteable phrasing.

Why this file is an accidental gift to coaches and scouts: the underlying structure of teaching transfers cleanly. A CS336 student has to write the code to actually learn programming, just as an athlete has to perform the movement to actually internalize the technique. A teacher uses guiding questions to help a student debug their reasoning, just as a coach uses "why did your inside foot stop there" to help an athlete unpack a turnover. Stanford has packaged the "AI training assistant equals teaching co-pilot, not replacement head coach" boundary into an open Markdown file that any coach can fork in ten minutes, swapping out "Python / PyTorch / bash" for "shooting form / footwork / practice plan" to get their own AI coaching assistant guidelines.

3. How Coaches and Scouts Use This: Five Steps to Adapt the Stanford Template Into AI Coaching Assistant Guidelines

The hard part of adopting AI coaching assistant guidelines is never the technology — it is the shared-boundary conversation. If you are a high-school coach, an AAU head coach, a community-club coach, or an independent scout, the five steps below can be run by anyone, no code required.

Step 1: Fork the CS336 CLAUDE.md as your skeleton. Copy the raw Markdown from stanford-cs336/assignment1-basics/CLAUDE.md, rename "CS336" to your team's name, and swap "Python / PyTorch / Triton" for "shooting form / defensive footwork / tactical spacing." This is a search-and-replace pass, not a coding task.

Step 2: Translate the "AI SHOULD" list into practice-floor actions. Examples that map cleanly: explaining why a drop step beats a spin move in the low post (allowed); pointing out that "the right leg moved first, the left leg did not catch up" in an athlete's training clip (allowed); citing specific paragraphs of the FIBA or NCAA rulebook (allowed); recommending five targeted drills aimed at a specific weakness the coach has flagged (allowed).

Step 3: Translate the "AI SHOULD NOT" list into hard red lines. Examples: do not write the full "here is how you should play tomorrow" pre-game plan for an athlete; do not tell an athlete "your move is wrong, here is the correct move"; do not deliver tactical adjustments directly to athletes without coach review; do not generate the lineup decision; do not give a scout the conclusion "this kid deserves a D1 scholarship." BLS explicitly lists "choose the appropriate players to use during a game" and "be selective when recruiting players" in the coach's core qualities — handing those calls to AI is handing away the core of the job.

Step 4: Rewrite the six "Teaching Approach" actions as a "Coaching Approach" six-step. Ask clarifying questions becomes "let the AI ask the athlete what their footwork was before the turnover." Reference concepts becomes "let the AI quote the coach's own previously published training manual." Suggest next steps becomes "let the AI suggest adding ten minutes of drop-step drill to the next practice." Review their code becomes "let the AI review the film clip and flag three specific improvement areas." Explain the why becomes "let the AI surface the mechanics or tactical reasoning behind each recommendation." Prefer tests and invariants over fixes becomes "let the AI propose measurable micro-drills rather than verdict-style conclusions."

Step 5: Drop the finished AI coaching assistant guidelines into the system prompt slot of every AI tool the team touches. That means pasting the Markdown into the Project knowledge of a Claude Project, the Instructions field of a ChatGPT Custom GPT, the system prompt of a Cursor profile, or the initial instruction of a local Ollama / LM Studio setup. Wherever an athlete, parent, or assistant coach chats with AI through a team-provided entry point, the output is now constrained by the same document.

4. Cases and Outcomes: Turning the "AI Is Replacing My Coaching Voice" Anxiety Into an AI Coaching Assistant Guidelines Dividend

Data shows that the moment AI coaching assistant guidelines are in place, three outcomes appear quickly: athlete learning depth recovers, scouting-report voice consolidates, and coach hours get freed.

Athlete learning depth recovers. When the AI training assistant follows the guideline of "ask first, guide second, let the athlete arrive at the conclusion last," athletes stop importing ChatGPT answers wholesale into practice. Research shows that athletes who actively self-review internalize technique 30–50% faster than athletes who passively accept answers. The "motivate, develop, and direct athletes" duty BLS calls out gets amplified in the AI era rather than diluted by it.

Scouting-report voice consolidates. Once an "AI scouting assistant guideline" sits as the shared system prompt, assistant scouts, volunteer parents, and the head coach all produce reports with the same structure, the same scoring axes, and the same compliance disclaimer language. That matters disproportionately to independent scouts — BLS data shows roughly 10% of coaches and scouts are self-employed, and what they sell is "trust this report."

Coach hours get freed. BLS notes coaches "may work more than 40 hours a week for several months during the sports season." A well-written set of AI coaching assistant guidelines lets AI safely absorb the pure-text overhead — game-film notes, weekly opponent briefs, parent-update emails, athlete-conditioning summaries — and hands the coach back hours for what BLS keeps insisting is the irreplaceable part: in-person instruction and decision-making.

5. FAQ: Five Questions Working Coaches Ask About AI Coaching Assistant Guidelines

Q1: Is the Stanford CS336 CLAUDE.md really open and copyable?

A1: Yes. The file lives in the public GitHub repository stanford-cs336/assignment1-basics, runs 74 lines and 4.74 KB, and under GitHub's default terms is free to fork, adapt, and reference commercially. Coaches only need to credit "adapted from Stanford CS336 CLAUDE.md" at the top of their AI coaching assistant guidelines.

Q2: BLS says coaching jobs grow 6% over ten years. Do I really need to care about AI coaching assistant guidelines now?

A2: According to U.S. Bureau of Labor Statistics August 2025 data, coaches and scouts employment is projected to grow 6% from 2024 to 2034 — above the 3% all-occupations average — with 41,800 annual openings. Competition is still tight. AI coaching assistant guidelines do not preserve "the job you already have"; they let one coach deliver what used to take two or three, which is how you reach the limited higher-paid seats and advancement tracks.

Q3: Most of my athletes' families aren't technical. How are they supposed to honor an AI coaching assistant guidelines document?

A3: You do not need the athletes to honor it. You need the AI tools your athletes use to honor it. Paste the guideline into the team-provided Claude Project or Custom GPT. When athletes use that entry point, the AI responds with "ask first, guide second, no direct answers" by construction. Parents don't need to understand the technology; they only need to ensure their kids ask through the team entry.

Q4: Scouting reports involve minors' video and personal data. How do AI coaching assistant guidelines handle privacy?

A4: Put "never upload identifiable minor-athlete video, names, addresses, or family information to any cloud AI" as the first item in the "SHOULD NOT" list of your guidelines. For sensitive content, prefer a local LLM (Ollama running Llama 3.1 8B, Gemma 3, or Qwen 3) so data stays on the coach's laptop or team NAS. Research shows local inference at the 8B parameter level already handles roughly 90% of youth-sports text workloads.

Q5: I'm not technical. Which of the five steps am I most likely to stall on?

A5: The hardest and most important step is Step 5 — getting the guideline into the system-prompt slot. The simplest entry points are Claude Projects (create a Project on claude.ai and paste the Markdown into "Project knowledge") or ChatGPT Custom GPTs (paste it into the Instructions field of GPT Builder). Both are no-code and take about ten minutes end-to-end.

6. Action Checklist: Three Steps to Stand Up Your AI Coaching Assistant Guidelines This Week

First, tonight, spend ten minutes opening Stanford CS336 CLAUDE.md and doing a search-and-replace pass: every instance of "code / Python / PyTorch" gets swapped for the equivalent concept in your sport. That produces your first draft AI coaching assistant guidelines.

Second, this week, create a dedicated Claude Project or ChatGPT Custom GPT, paste the draft into the system prompt, and invite a single AI-savvy starter to stress-test it. Watch whether the AI actually asks back, guides instead of answering, and refuses to hand over conclusions.

Third, at next week's pre-practice meeting, print the AI coaching assistant guidelines and hand them out to the full roster. Spell out the difference between the in-team AI entry point and the outside ChatGPT/Claude/Gemini entry points. Tell athletes why the in-team AI will not just hand them the answer — because that answer is the growth path you, as their coach, want them to walk themselves.

Coaches and scouts are not a profession AI replaces; they are one of the earliest professions that can turn AI into a co-pilot on the training floor. BLS has already put the numbers on the table — 306,500 jobs, 6% growth, $45,920 median pay. Stanford CS336's June 2026 CLAUDE.md hands every high-school, AAU, and community-club coach a ready-made AI coaching assistant guidelines template. The remaining question is whether you fork it tonight, or let your athletes spend another week chatting with unconstrained AI.