Winning by Design / Course Review

AI for GTM: Pilot Cohort Debrief and the Path to Three Tiers

What landed, what to fix, and how to split one course into Starter, Intermediate, and Advanced, grounded in the recordings.

Cohort: Charter pilot, week of July 20, 2026 | Learners: ~22, mostly Growth Institute members | Facilitator: Dan Smith
Reviewed: 4 sessions across 5 recordings (S1 local, S2 to S4 in Grain), every claim below tied to a transcript quote.

The one thing to take away

The "this feels too basic" signal from the room was not a content problem. It was an audience-routing problem. You built a Level 2 course for frontline ICs, then filled the pilot with a Level 3 audience who had already built the tools you were teaching. The fix is not to rewrite the content. It is to run three tiers and route people to the right one. The room spec'd this for you: "AI for ICs, AI for managers, and AI for GTM leaders."

01Scorecard

Strong pilotship the three-tier split

A genuinely strong first run of a brand-new format. The craft is there and the demand is proven. The gaps are routing and logistics, not teaching: a Level 3 audience got Level 2 content, a mixed-platform room got Claude-only material, and setup friction hit the core deliverables. All fixable before August.

Dimension	Rating	What the recordings show
Facilitation and inclusion	Strong	Tracks each person by name, converts anecdotes into principles, handles pushback with grace. Shari: "they appreciate your style of including them."
Adaptiveness across the week	Strong	Diagnosed his own lecture bias and corrected it session by session, in real time.
Build vs lecture ratio	Mixed	Hands-on build swung from ~5% (S1) to 0% (S2, breakout cut) to ~38% (S4). Trending right, not yet steady.
Audience-to-tier fit	Off	A Level 3 room of leaders and consultants sat through Level 2 IC content. The dominant signal of the week.
Single-platform fit	Off	Claude-centric teaching with ChatGPT users in the seats. Olivier was "lost in the vocabulary."
Tech and setup readiness	Weak	LMS email gating, corporate connector approvals, a Docebo-to-API bug. It hit the core deliverables.
Content depth for the target IC	Strong	Right altitude for the intended frontline rep, even though this particular room was already past it.
Certification clarity	Needs work	Learners were unsure what to submit. Alejandro: "what should I deliver for the very last one, the final exam?"

02Key timeline: the lecture-to-breakout swing

Approximate session shapes, reconstructed from the transcripts (not minute-exact). The story is the arc: you caught your own lecture bias mid-week and corrected it. Session 2 is the dip, where the planned breakout got cut entirely.

Welcome / check-in Lecture / teach Breakout build Debrief Q&A / discussion

S1Jul 20

~5%

S2Jul 21

S3Jul 22

~30%

S4Jul 23

~38%

~5%

S1 hands-on build

S2 (breakout cut)

~30%

S3 hands-on build

~38%

S4 hands-on build

Roughly two thirds of a skills workshop should be participants building or being debriefed. You started well below that line and, by Session 4, were nearly on it. Locking a protected 20-minute build into every session makes the good end of this arc the default rather than the finale.

03Key moments to watch

Nine moments worth a highlight reel: six learner reactions and three points that landed with the room. Each card jumps to that moment in the recording. These are moment cards with deep links, not video stills, since the Grain transcripts do not include frames.

S215:00

Landed strong

"The top-performing teams put AI to work while they are asleep. Most of us are still using it to write a better email."

"We gained one day per person per week by building a prospecting research agent."

"To me it is kind of a dream. I do not want to be too dramatic, but that is really cool."

"Unless it is adopted, it is about as useful as a chocolate teapot."

"When reps can self-coach, that is the primary reason we have middle management today."

"I have one skill that reaches out to five connectors, runs four transformations, and sends the finished product out through Slack."

"It brought the inconsistencies up as a discussion point. That is pretty impressive for an AI."

"Treat it like a brilliant intern that does not yet know what to do. Smart, but you guide it."

"A program of extremely great value to me."

Valdemar Duarte

Watch this moment

04What you did well

These are your moat. They are also exactly what Winning by Design is known for, and the pilot proved you still have it in a brand new format.

Facilitation and inclusion

You open by tracking each person individually, not with a syllabus. "Olga, I saw your name pop up in the chat that you were engaging in convo with some other Growth Institute members."
You convert learner anecdotes into transferable principles on the fly: "The static dashboard is gone. What you are implying is almost this interactive dashboard."
Learners feel it. Shari, in the debrief: "they appreciate your style of including them in the conversation."

Adaptiveness across the week

You diagnosed your own lecture-heavy tendency in real time and corrected it day by day. After S2: "I swung the pendulum in session 2 too far towards lecture."
By S3 you shipped a real 20 minute build plus a 14 minute debrief. By S4: "I love the pivot to more of the breakout rooms. That is something Winning by Design is known for."
When Graydon asked for more peer building, you committed on the spot: "I will reprioritize that for tomorrow and readjust my slides and timing."

Outcome-first, build-not-watch pedagogy

The arc worked. Lou: "You saw session 1, 2, 3, 4, all trying to build on each other. Instead of reading or watching a video, it is go build something and have it iterate with you."
Signature reusable techniques stuck: "ask me one question at a time," "treat it like an overeager intern," "cite where it got that information."

Credibility and humility

You de-shame the learning curve with your own struggle: "it took him, as a real AI expert, 8 iterations to get it right."
You anchor everything in real proof, not theory: the SPICED uplift "off of Salesforce fields across 50,000 opportunities," the ERP roadmap deep-research story, your own SPICED Pulse.
Strong testimonials landed. Valdemar: "a program of extremely great value to me." Graydon: "if I was still an AE, I would be chomping at the bit."

05Session by session

S1Your Automated Daily Brief~63 min / Jul 20

Worked

Opened with an insight hook, not a syllabus, then a breakout in the first minutes to build community.
Gave the cohort a shared north-star vocabulary: "euphoric surprise."
Handled a sharp challenge well. Louise: "the competitive advantage is not with reps, it is with the managers." You validated it fully.

To improve

For a session titled "build your daily brief," nobody built one. The actual build was pushed to homework, high risk busy ICs never do it.
All the friction hit at the end, unprepared for: LMS email gating blocked homework submission, and Angela had to fall back to Copilot on IT approval grounds.
Model-selection content drifted into your own frontier setup (Fable, Claude Code, off-hours token economics), above where most ICs live.

S2Research While You Sleep, Sell While You're Awake~70 min / Jul 21

Worked

A genuinely useful, counterintuitive core lesson: do not craft the perfect prompt, have AI write the prompt for you.
You invited dissent and elevated it. Harry delivered the ROI moment of the week: "we gained 1 day per person per week through a prospecting research agent."
You taught weighting and ranking of research by critical-event urgency, a real intermediate skill.

To improve

You ran behind and cut the single planned interactive moment: "I am a touch behind, so I am gonna skip actually doing one of these things live." A session about running a prompt had no one run a prompt.
Effectively no Q&A: "I have a minute exactly left." The best exchanges only happened because people stayed after.
Between-session assignment had no crisp deliverable: "try and build out what an overnight run would look like," no template, no success test.

S3Run the Call, Prove the Impact~76 min / Jul 22

Worked

You delivered the "heavier breakout" correction: a real 20 minute build with protected individual work, "build individually, but build it as a crew."
Reduced the hardest SPICED elements to buildable definitions: "critical event is simply the date the customer needs to achieve that impact by."
Owned a live mistake cleanly: "I moved too fast and I did not validate it, so I am gonna do a quick prompt update."

To improve

The breakout offered four different prompt paths, so no shared artifact emerged. Marcus had to ask: "are there 4 breakout rooms, or does everyone do whatever they want individually?"
The prompts-need-context critique went unresolved. Louis: "they just read off a list, and that becomes next to useless." You deferred it to the next session.
The richest strategic material (inspection versus enablement, middle-management disruption) happened entirely in overtime with 2 to 3 people, so most of the cohort missed it.

S4Turn Every Call Into a Repeatable Skill~52 min across two parts / Jul 23

Worked

Best breakout balance of the week, and a clean 5-part mental model for building a skill (invocation, inputs, constraints, output, citations).
You crowdsourced the definition rather than lecturing it, and let Graydon and Louis co-teach "skill versus workflow."
You closed by mining the room for the tiering and role-lens insights that now drive this whole review.

To improve

The breakout was over-scoped: "we cannot probably complete all of these 5 steps in a 20-minute breakout." You changed the example last minute, and it "did not land the same for everybody."
Certification was fuzzy in the moment. Alejandro: "I just want to understand what I should deliver for the very last one, the final exam."
ChatGPT users were lost on Claude-only vocabulary. Olivier: "I am a bit lost in the vocabulary. I am using ChatGPT. The notion of skills is not even known to me."

06Cross-session patterns

These repeat across all four sessions. They are the real agenda for the next cohort.

Pattern	What the recordings show
Audience and tier mismatch The dominant signal	A Level 3 room got Level 2 content. Louis, after S3: "most of the people on the call are probably doing a lot of this stuff already." The advanced learners had already built the exact tools (Graydon, Olivier). You named it every session but could not fix it live, because it is a routing problem, not a content problem.
The lecture-to-breakout swing	A visible self-correction arc: one 2 to 3 minute breakout in S1, breakout cut entirely in S2, a real 20 minute build in S3, best balance in S4. Your own repeated diagnosis is the evidence: "I tend to monologue," then "I am doing the thing, talking fast, monologuing." The skill is clearly trending the right way.
Mixed platform in one room	Claude-centric teaching, but ChatGPT users in the seats. Valdemar mapped "skills" to custom GPTs to keep up; Olivier was simply lost. Louis switched to ChatGPT mid-breakout "because it uses so many less credits." One platform per cohort would remove this entirely.
Setup and connector friction	Concentrated but recurring: LMS email gating (S1), a Docebo to Claude API permissions carryover bug (S2), voice mode not set up for all (S3). Your own S4 conclusion: "make connectors a prerequisite by session 4." Selena: "pre-work where we all set up our connectors and MCP before the start."
Breakout deliverable not crisp	When the breakout has one shared artifact it sings; when it forks (four prompt paths in S3, five steps in S4) it confuses and produces nothing shared. Marcus and others had to ask for basic structure.
Prompts are starting points, not answers	The strongest content critique, raised repeatedly and not yet resolved. Louis: prompts "are fantastic starting points, but they are starting points." Graydon: the tool "never helps the rep produce a different behavior, it is no substitute for coaching." Teaching prompt plus context plus iteration explicitly would close this.
Cross-company exercise friction	A mixed open cohort cannot share real deal data. Alejandro: "we should create a fictitious environment, a CRM system." Your own answer is the build: 100 synthetic transcripts and a mock CRM via a custom MCP.
Role-lens demand	The same exercise wants a different lens per seat. Selena: "run it through a different lens of the role depending on where you are in the bowtie." This is breakout-level personalization, not a separate course.

07The course architecture

Your instinct is right, and the pilot is the proof. The current iteration is Level 2. The room told you what Level 1 and Level 3 are.

Why the split solves two problems at once

Proficiency tier and platform are two different axes, and one enrollment router fixes both. Ask two questions at signup: which assistant does your company use, and how would you rate your AI proficiency. Route to a single-platform cohort at the right tier. The planning doc already specified "the enrollment form always asks which assistant, we route, we never assume." The pilot simply did not enforce it. The interactive AI-journey router does exactly this live: a few questions sort each person to Starter, Intermediate, Advanced, or the Executive track.

Level 1

Starter

Who

Frontline AEs, SDRs, CSMs with a license they barely use. The stated target that was NOT in the pilot room.

Prereq

A working license and LMS access. No connectors required.

Outcomes

AI literacy (models, context, "one question at a time," never trust the first output), the daily brief, basic research, call prep with provided prompts.

Format

Heavy in-session guided builds. Everyone ships the artifact live, web-only, no wiring.

Moves down from now

The S1 daily-brief build, "ask AI to write the prompt," cite-sources, basic model selection.

Level 2

Intermediate

This iteration

Who

Reps already comfortable in chat who want real workflow leverage.

Prereq

Starter or equivalent, plus connectors set up before day 1 (hard gate).

Outcomes

Overnight deep-research agents, SPICED-scored call prep and post-call review, connectors and MCP wired in, one reusable skill shipped.

Format

The 4-session build arc you already ran. Keep it.

What it fixes vs pilot

Single platform per cohort, connector pre-flight gate, one crisp breakout artifact, resolve the prompt-context gap.

Level 3

Advanced: Builder

Who

Power users who build for the team. The Graydon tier: already running skills and agents in production.

Prereq

Fluent daily user, comfortable with connectors and the command line.

Outcomes

Team skill libraries in a GitHub repo or Claude Enterprise, multi-agent workflows and orchestration, self-learning coaching loops, SPICED Pulse-style dashboards.

Format

Hands-on builder track. Ship reusable infrastructure the whole team runs on.

Source of the spec

Graydon in the room: one skill that "reaches out to 5 connectors, performs 4 transformations, and sends the finished product out through Slack," plus his org's shared repo everyone builds from.

Separate track

Executive

Workshop

Who

GTM leaders: CROs, VPs, RevOps. A different axis, not a rung on the ladder.

Prereq

You lead or shape a GTM team.

Outcomes

A maturity read on your org, the three shifts, one motion redesigned agent-first, and a 90-day commitment.

Format

A 2-hour strategic workshop, not a prompt class. The run of show is built.

Source of the spec

The room: "AI for managers, AI for GTM leaders," and "the CRO perspective, architecting the whole platform."

The nuance worth holding: the pilot felt "too basic" precisely because a Level 3 audience sat through Level 2 material, and the Level 3 content they actually wanted only surfaced in overtime with two or three people. That is not a knock on the course. It is the clearest possible market signal that Level 3 should exist and sells to the exact people already in your Growth Institute orbit.

Platform is the second axis, role is the third

Keep it simple to communicate: lead with the three proficiency tiers. Platform (Claude master, Copilot and OpenAI as ports of the same spine) is a parallel track already in the plan, and cohorts never mix. Role lens (Prospecting, Selling, CS, AM) is not a fourth course, it is how each breakout personalizes, via a self-select prompt, the way Selena described.

L2 S1

Daily brief and prioritization routine

L2 S2

Overnight deep research and pre-call brief

L2 S3

SPICED-scored discovery and post-call review

L2 S4

Package it as one reusable skill

The current 4-session arc becomes the Level 2 spine. Starter is a gentler on-ramp to it; Advanced is what comes after.

08Fix these before the August open cohort

Distinct from the strategic tiering. These are the operational fixes the recordings demand, most were even specified in your own plan and just need enforcing.

Hard pre-flight setup gate at T-3.Verify LMS email equals enrollment email, confirm the IT-approved LLM, send a connector permission checklist to the learner AND their IT contact. This kills the S1 email-gating and Copilot-fallback failures before they happen.
One platform per cohort, routed at enrollment.Stop mixing Claude and ChatGPT in one room. It is the single cheapest fix for the vocabulary confusion and the credit-cost anxiety.
One crisp shared breakout deliverable per session.Cut the choose-your-own-adventure sprawl. Protect a full 20 minutes. Everyone leaves the breakout with the same artifact built on their own work.
Pre-build a fictitious org, mock CRM, and synthetic transcripts.Your own idea. It lets a mixed open cohort do real exercises without exposing confidential deal data. Alejandro asked for exactly this.
Teach prompt plus context plus iteration, not prompt libraries.Directly answers Louis and Graydon. Add the "show it what good looks like" and "feed it your own examples" step so outputs stop sounding like AI slop.
State the certification deliverable and rubric on day 1.The final is the skill. Say so up front, show the rubric, so no one is asking "what do I submit" in the last five minutes.
Run the comms cadence you already designed.Slides, Slack link, and a recap within 24 hours of each session. The plan specifies it; the pilot let a couple of these slip.
Set the week's expectations and success criteria on day 1.Add the "what to expect" slide the debrief already flagged. Name the arc, what "done" looks like, and how homework gets reviewed.

09Bottom line

The pilot worked. Your facilitation, your adaptiveness across the four days, and the build-not-watch format are genuinely strong and hard to copy. The feedback that stung, "this feels too basic," is the best possible news: it means the demand for a harder tier is already sitting in your Growth Institute audience, and they described it to you in their own words. Split the course into Starter, Intermediate, and Advanced, keep this iteration as the Level 2 spine, route people by platform and proficiency at enrollment, and fix the eight operational items above before August. The content is not the problem. The routing is.

Prepared for Dan Smith. Every quote is drawn from the S1 to S4 recordings and the July 20 debrief. S1 transcribed locally from the Part 1 MP4; S2 to S4 from Grain. Where the diarizer mislabeled the facilitator, teaching content is attributed to Dan.