AI in GTM School · Q2 2026
Class 2 of 8 · Recap & Takeaways

Growth Strategy Design for an Enterprise

Personal AI fluency, the Master Context Doc, and the move from one-off prompts to scheduled automation — Ryan Staley's working session for ~260 GTM operators.

Instructor: Ryan Staley (Whale Boss)
Co-host: Andy Jolls
Date: May 6, 2026
Cohort: ~260 attendees

TL;DR

Today's class shifted gears from theory to hands-on. Ryan walked the cohort through building a Master Context Doc — your company, ICP, value prop, case studies — that becomes the spine every Claude session reads from. The move he kept driving home: stop typing one-off prompts. Build a personal context layer once, then use Claude Code's plan mode to design repeatable workflows. Pair that with the right model for the job (Opus to plan, Haiku to execute) and the unit economics of AI in your week change overnight.

The session arc

Three concepts, in order

Ryan refined this curriculum across 3,000+ operators over three years. The order matters — fluency is the prerequisite to cascade is the prerequisite to apply.

1

Personal AI Fluency

You can't lead a team through AI if you can't run your own desk with it. Step one is building a personal context layer the model can read every session — your role, your ICP, your competitors, your differentiator. Without it, every prompt starts at zero.

"Don't multitask. This is going to seem like it's going at breakneck speed. If you're trying to type emails or do other stuff, you're going to miss out on a ton."
2

Cascade

Once your own workflow runs on AI, the next move is to cascade it — to your team, to the function, to the company. The Master Context Doc isn't just yours. It becomes the shared spine your VPs and managers point their own Claude sessions at, so every output speaks the same brand voice.

3

Apply

Fluency + cascade only matter if they show up in pipeline. The hands-on portion focused on applying the system to real GTM jobs: daily planning automation, outbound research, deal strategy. The principle: pick one workflow, automate it end-to-end, then move to the next.

The Cohort Poll

Where the room sits on tools

Quick start-of-class poll on primary model. Compared to the January cohort, Claude's dominance has grown — and ChatGPT's primary-tool share has compressed dramatically.

Primary model Q2 2026 cohort Where it wins Ryan's take
Claude 80% Long-form reasoning, agentic work, Cowork's structured projects, Claude Code for terminal-driven automation The default for almost everyone in the room. Pair Opus (plan) with Haiku (execute) for cost.
ChatGPT 10% Image generation, voice mode, broadest plugin ecosystem Still useful but the cohort has shifted. Use it for what it's best at, not as the default.
Gemini 6% Google Workspace integration, long-context document analysis, NotebookLM "Underused. If you live in Workspace, the security envelope alone is worth it."
Copilot / others ~4% Microsoft 365 enterprises with data-privacy constraints Default if you're already a Microsoft shop — Copilot has Claude under the hood.

The Centerpiece

The Master Context Doc

A single markdown file that travels into every Claude session and tells the model who you are, what you sell, and how you sell it. Ryan walked through it live; here's the structure.

1

About You

Role, company, what you actually do day-to-day. Three sentences. The model doesn't need a resume — it needs the lens.

2

What You Sell

One paragraph product description. Target buyer title. Average deal size. Sales cycle length. Pricing model. Anything that anchors the model in your real economics.

3

ICP & Anti-ICP

Industries, company size, the signals that mean "good fit." Equally important: who's not your ICP and why. The model needs both halves to triage effectively.

4

Competitors & Positioning

Top 3–5 competitors, one sentence each on how you position against them. This is the layer that makes the difference between generic AI output and feels-like-our-team output.

5

Case Studies & Outcomes

Two or three real customer outcomes — the named account, the problem, the numerical result. The model uses these as anchors for tone, depth, and specificity.

6

Tools You Use

CRM, outreach platform, enrichment, conversation intelligence. The model needs to know what it can suggest and what it can't.

"Pinning a Master Context Doc into Claude is the difference between writing a prompt and running a system. You write it once. Every chat after that is downstream of that one file."— Ryan Staley

Hands-on

Two workflows the room built live

Ryan ran two live builds — both starting from the Master Context Doc, both using Claude Code in plan mode.

Workflow 1

Daily planning automation

  • Trigger: a single chat command at the start of each day
  • Reads calendar + last 24h of emails + open CRM tasks
  • Returns prioritized list framed around your real quota and pipeline
  • Built in Claude Code's plan mode first, then activated as a daily task
Takeaway: The model isn't picking your day — it's doing the triage you would have done in 20 minutes, in 20 seconds, and pulling signal from sources you'd never read by hand.
Workflow 2

Outbound research + deal strategy

  • Master Context Doc supplies ICP + competitor framing
  • Cursor + Exa for LinkedIn pulls (no firewall friction)
  • Apollo or Clay for contact enrichment if the security review allows it
  • Output: structured account brief + competitive positioning + first-touch message
Takeaway: The outbound stack is composable. Pick the tools your security team has already cleared and chain them; don't fight to add a new vendor mid-workflow.

Model Economics

Opus to plan, Haiku to execute

A cohort question opened up the most concrete cost lever Ryan surfaced all session.

"I do the planning in Opus, but I'm actually using Haiku to do the work — because most of it doesn't need a heavy lift. Haiku is about a tenth of the price. So a lot less token intensive."— Andy Jolls, riffing on Ryan's pattern
Plan in Opus. When the work is novel — designing the workflow, debugging an edge case, writing the first version of a skill — pay for the bigger model. You're paying for one good answer, not a hundred.
Execute in Haiku. Once the pattern is clear, the heavy lifting is repetition — and Haiku is ~10× cheaper per token. The math compounds fast on any workflow you run daily.
Watch for credit burn. Ryan flagged a CEO automation eating $25 of API credits every few days. The fix: route the heavy-volume legs to a local or open-source model.
Open-source = future-state. Most people in the room aren't there yet. Get to Haiku-for-execution first; revisit open-source when you start hitting API credit ceilings.

Sticky Quotes

Lines worth saving

"Don't multitask. This is going to seem like it's going at breakneck speed. If you want to get something out of this, you have to be present."— Ryan Staley
"Your fear is going to go through the floor. Your confidence is going to go through the roof. That's really what I want for y'all because this is happening so fast."— Ryan Staley
"Once you have the plan dialed in, you say go — and it'll create it. It doesn't mean it's going to automatically work, but the plan is where you measure twice and cut once."— Ryan Staley, on Claude Code's plan mode
"If you have a note taker on, please kill it. It's very distracting. I want to look at your faces."— Ryan Staley, setting the room

The Cohort Asked

Questions worth the replay

The five most-engaged threads

  • "For enterprise deployment, how do I make sure every user's Claude session references central context — ICP, competitors, brand voice?" The Master Context Doc is the answer; the cascading variant is to publish a "master" Master Context Doc your team forks into their own Claude.
  • "Sonnet vs. Opus — does it actually change the output on this work?" Yes for novel design; less for repetitive execution. Default to Sonnet for cost; reach for Opus when the work is genuinely new.
  • "How do I migrate between Claude instances when I move from personal to company?" Treat the Master Context Doc as your portable layer. Files travel; chat history doesn't.
  • "How do I connect this to outbound automation without spamming?" Cursor + Exa pulls LinkedIn without firewall friction; Apollo or Clay if security has cleared them. The model still drafts the personalization — you ship volume only after the message quality holds.
  • "Should we be looking at open-source models yet?" Only if you're hitting credit ceilings on production automations. Get to Haiku-for-execution first.

Before Class 3

What to do this week

Ryan's homework for the room — concrete, finite, and doable in a single Sunday afternoon.

Refine your Master Context Doc. Six sections: About You, What You Sell, ICP & Anti-ICP, Competitors, Case Studies, Tools. One page. Save it as CLAUDE.md if you're in Claude Code; pin it as a Project in Cowork.
Pick one daily workflow to automate. Daily planning is the lowest-friction starter. Ask Claude in plan mode: "Here's how I plan my day today. How could you do this for me?" Iterate the plan, then run it.
Test Opus → Haiku routing. For one workflow this week, plan in Opus and execute in Haiku. Note the cost delta and the quality delta. Most of the time the quality delta is rounding-error.
Bring one workflow to your cohort group. Don't workshop it alone. The cohort breakouts are where this content compounds; show what you built and steal what they built.