gtm-ai-gtm

SKILL.md

AI Product GTM

Go-to-market strategy for AI products. These aren't generic AI principles — they're patterns from selling autonomous AI agents into enterprises where "autonomous" scared buyers and "teammate" converted them.

When to Use

Triggers:

"How do we position this AI product?"

"Buyers say they're worried about AI breaking production"

"Should we call it autonomous or copilot?"

"How do we price AI when usage varies 10x by customer?"

"Enterprise security passed but ops rejected us — why?"

Context:

AI agent platforms (coding, support, ops)

LLM-based applications

Autonomous tools that do things (not just suggest)

AI infrastructure

Anything where the AI makes decisions

Core Frameworks

1. The Real Enterprise AI Objection (It's Not What You Think)

What I Learned Selling Autonomous AI Agents:

Three months in, enterprise security reviews were passing fast. Good sign, right? Then the pattern emerged: security approved, but operations rejected us.

The objection wasn't "will the AI break production?" — they assumed it would break production eventually. The real question was:

"Who's responsible when the agent does something wrong?"

Not "do we trust the agent?" — "do we trust our team to handle this?"

Why This Matters:

Autonomous agents create a new operational burden. You're not selling AI capability, you're selling organizational readiness. When your agent halts production at 2am, who gets paged? Who fixes it? Who explains it to the VP?

Framework: The Accountability Cascade

Before deploying AI agents, enterprises need clear answers:

L1 Response: Who monitors the agent? (24/7 ops team, or dev team on-call?)

L2 Escalation: When agent action fails, who debugs? (Agent team, or product team?)

L3 Ownership: When something breaks badly, who owns customer communication?

If you can't answer all three, they won't buy. Doesn't matter how good your AI is.

How This Changes Your Sales Process:

Old approach:

Demo the AI

Show accuracy metrics

Talk about ROI

New approach:

Demo the AI

Show the failure modes explicitly

Ask: "Who on your team would handle this scenario?"

Walk through their incident response process

Map AI failures to their existing runbooks

The Qualification Question:

"Walk me through what happens when the agent takes an action that breaks a workflow. Who gets alerted? Who investigates? Who decides whether to roll back or fix forward?"

If they can't answer, they're not ready. Pause the deal and help them build the process first.

Common Mistake:

Treating this as a product objection ("we'll make the AI more accurate"). It's an organizational objection. More accuracy doesn't solve "who owns this at 2am?"

Pattern I've Seen Work:

Companies that succeed with AI agents already have:

On-call rotations for production systems

Incident response playbooks

Blameless postmortem culture

Clear escalation paths

Companies that struggle:

Manual deployment processes

Hero culture ("Steve fixes everything")

No formal incident response

Blame-focused culture

Decision Criteria:

Before demoing autonomous AI to enterprises, ask yourself: "If this breaks their production, who on their team owns the fix?" If you can't answer, they can't buy.

2. Copilot vs Agent vs Teammate (Three Different GTM Motions)

The Positioning Trap:

Early enterprise conversations, we positioned as "autonomous AI agent." Buyers flinched. One word change — "autonomous" → "AI teammate" — and deal progression improved measurably.

Why? Word choice shapes buyer psychology.

The Three Framings:

1. Copilot (Safest, Lowest Value)

What it means: AI suggests, human decides every time

Buyer psychology: Feels safe, non-threatening

GTM motion: Developer adoption, bottoms-up

Use case: Code completion, writing assistance, search

Objection: "Is this worth paying for?" (low perceived value)

2. Agent (Scariest, Highest Value)

What it means: AI acts autonomously, human reviews periodically

Buyer psychology: Scary, implies replacing humans

GTM motion: Enterprise sales, top-down

Use case: Batch processing, automated workflows, ops

Objection: "What if it breaks production?" (accountability fear)

3. Teammate (Sweet Spot)

What it means: AI and human collaborate, split the work

Buyer psychology: Partnership, not replacement

GTM motion: Hybrid (dev adoption + manager approval)

Use case: Most AI agent platforms

Objection: "How do we integrate this into our workflow?" (process question)

The Positioning Shift:

Before: "Autonomous AI agent that handles complex workflows end-to-end"

Developers: "Cool, but scary"

Managers: "Will this replace our team?"

Deal progression: Slow

After: "AI teammate that pairs with your engineers on complex tasks"

Developers: "This helps me"

Managers: "This makes my team more productive"

Deal progression: Three enterprise deals that had stalled 4+ months closed within 8 weeks of the shift

Specific Language Choices That Mattered:

❌ Don't say:

"Autonomous" (scary)

"Replaces" (threatening)

"Fully automated" (no control)

"AI-first" (what does that even mean?)

✅ Do say:

"Teammate" (collaborative)

"Augments" or "Accelerates" (helping, not replacing)

"You stay in control" (reassuring)

"Handles the repetitive work" (specific value)

How to Choose Your Framing:

Does your AI make decisions without human approval?
├─ Yes → Are you selling to developers or enterprises?
│   ├─ Developers → "Agent" framing (they want autonomous)
│   └─ Enterprises → "Teammate" framing (they want control)
└─ No → "Copilot" framing (augmentation, not automation)

The Hard Truth:

You can build an agent but position it as a copilot. You can't build a copilot and position it as an agent. Product capabilities set a ceiling, positioning chooses where you land below it.

Common Mistake:

Using "autonomous" because it sounds impressive. Impressive ≠ trusted. If buyers flinch at your positioning, you've lost them.

3. The AI Pricing Problem (When Usage Varies 10x)

The Pattern:

Every AI company I've worked with faces this: Customer A uses 1,000 API calls/month. Customer B uses 10,000. Do you charge Customer B 10x more? If yes, they churn. If no, your margins collapse.

The Three Models:

1. Seat-Based ($X per user/month)

When it works: AI augments human work predictably

Example: Code completion, writing assistant

Problem: Doesn't capture AI value scaling

Real risk: High-usage customers are your best customers, but they subsidize low-usage ones

2. Usage-Based ($X per API call / prediction / hour)

When it works: AI does variable work, customers understand the unit

Example: Image generation, transcription, batch ML

Problem: Sticker shock for high-usage customers

Real risk: Customers optimize to use less of your product

3. Outcome-Based ($X per outcome achieved)

When it works: You can measure outcomes reliably

Example: "Pay per bug fixed" or "Pay per support ticket resolved"

Problem: Hard to measure, easy to game

Real risk: You bear all the risk if AI doesn't perform

What Actually Works (Hybrid):

Base fee (covers fixed costs) + variable fee (scales with value).

Example structure:

Base: $X/month per team (access to platform)

Variable: $Y per successful action/outcome

Why this works:

Base covers infra/support costs

Variable aligns with customer value

High-usage customers aren't punished (they're getting more value)

The Pricing Conversation I Wish I'd Had Earlier:

When pricing usage-based AI:

Ask the customer:
"How much would it cost you to do this manually?"

If it's $0.10 per API call but saves them $2 in labor, you're underpriced. If it costs $0.50 per call but saves them $0.40, they won't use it enough to matter.

Pricing Rule:

Your variable cost should be 20-30% of customer's alternative cost. High enough to capture value, low enough that they'll use it liberally.

Common Mistake:

Copying OpenAI's pricing ($0.01 per 1K tokens) because "that's what everyone does." Your cost structure isn't OpenAI's cost structure. Your value isn't OpenAI's value. Price for your business.

4. The AI Trust Ladder (From Someone Who Climbed It)

The Pattern:

You can't sell AI by saying "trust us, it works." You build trust in stages.

First: Transparency (Before First Demo)

Send these three docs before they ask:

Model card (what model, trained on what, accuracy on what benchmarks)

Security whitepaper (where data goes, how it's processed)

Explainability doc (how to interpret AI decisions)

Why this works:
Buyers expect to do diligence. If you send docs before they ask, you look confident and credible.

Second: Control (In the Demo)

Show them the safety mechanisms:

How users approve/reject AI suggestions

Kill switches and rollback mechanisms

Confidence scores and when AI says "I'm not sure"

Why this works:
Fear of "runaway AI" is real. Showing control mechanisms proves you thought about failure modes.

Third: Performance (Week 4-8)

Prove it works:

Benchmarks vs baseline (human or previous tool)

Case study from similar company

Live demo on their data (if possible)

Why this works:
Proof beats promises. One customer saying "we saved X hours/week" is worth 100 marketing claims.

Fourth: Scale (When They're Serious)

Show enterprise readiness:

Enterprise deployment examples

Performance at scale (latency, throughput, error rates)

Compliance docs (SOC 2, GDPR, etc.)

Why this works:
Enterprises don't deploy MVPs. They need proof you won't fall over at 1000 users.

The Mistake I Made:

Trying to prove performance before explaining how the AI worked. Buyers didn't trust the benchmarks because they didn't understand the system. Order matters.

Decision Criteria:

If buyers ask "how does this work?" before you've demoed, you skipped transparency. Back up and send the docs.

5. The Enterprise AI Demo (Show Failure, Not Just Success)

What Doesn't Work:

Canned demo where AI magically solves everything. Buyers think "this won't work on our messy data."

What Works:

Show the AI making a mistake and recovering. Seriously.

Demo Structure That Works:

1. The Problem (30 seconds)
"Your engineers spend hours on [specific task]. Here's what that looks like."

Show: Current manual workflow

Quantify: Time × Engineers × Weeks = Total cost

2. The AI Attempt (60 seconds)
"Here's the AI handling the same task."

Show: AI analyzing, taking action

Key move: Have AI encounter an error or uncertainty

Show: AI re-analyzing, recovering, or asking for help

Narrate: "Notice it didn't get it perfect first time. It handles uncertainty like a human would."

3. The Human Review (30 seconds)
"Here's where the engineer reviews and approves."

Show: Engineer examining AI's work

Key move: Show the engineer overriding or adjusting something

Narrate: "Human stays in control. AI handles repetitive work, human handles judgment calls."

4. The Outcome (30 seconds)
"[X hours] → [Y minutes]. Engineer still owns the outcome, AI accelerates execution."

Quantify: Time reduction, cost savings, capacity freed

Why This Works:

Showing failure → Builds trust (you're not hiding anything)

Showing recovery → Proves AI is robust

Showing human override → Gives them control

Quantifying savings → Makes ROI concrete

The Pattern I've Seen:

Demos with perfect AI → Buyers skeptical
Demos with imperfect AI that recovers → Buyers engaged

Common Mistake:

Cherry-picking examples where AI is 100% accurate. Buyers know real-world data is messy. If you don't show messiness, they assume you're hiding it.

6. The "Who Owns This?" Objection Handler

The Objection:

"This looks great, but what happens when the AI does something wrong?"

Bad Answer:
"Our AI is 95% accurate, and we're improving it every week."
(Translation: "It will break production 5% of the time, good luck with that")

Good Answer:
"Great question. Let's walk through a failure scenario together."

Then Ask:

"When the AI takes an action that causes an error, who on your team investigates?"

"Do you have an incident response process for tooling failures?"

"Who owns rollback decisions — the engineer who approved it, or the ops team?"

What This Does:

Shifts from "will it fail?" (yes, it will) to "how do we handle failures?"

Makes them think through operational readiness

Reveals whether they're ready for AI agents

The Follow-Up:

"Here's what we recommend: Start with low-risk environments. Let the AI handle non-critical workflows for 2-4 weeks. See how your team handles its mistakes. Then expand scope when you're confident in the process."

Why This Works:

You're not selling perfection. You're selling a tool that requires operational maturity. Filtering for mature buyers is better than convincing immature ones.

The Pattern:

Mature buyers say: "We already have runbooks for tool failures, we'll add AI to them."
Immature buyers say: "Can you make it never fail?"

Decision Criteria:

If a buyer demands 100% accuracy, walk away. They're not ready. Come back when they have incident response processes.

7. The AI Positioning Trap (Fighting Asymmetric Wars)

The Pattern:

You're competing in the AI agent space. Every competitor's homepage says the same thing: "Automate [workflow] with AI." Your differentiation requires explaining complex technical benchmarks that buyers don't understand.

This is the positioning trap: competing on features against better-funded companies on their battlefield.

How to Diagnose It:

Collect homepage messaging from 5-7 direct competitors

Identify shared claims (these are commoditized — you can't win here)

Map where you have structural advantage (not just product features)

Find the position competitors can't easily copy

Structural advantages that work for AI positioning:

Unique data or workflow ownership (you control something competitors can't replicate)

Deployment flexibility (on-prem, private cloud, customer-controlled infrastructure)

Pricing model innovation (outcome-based, usage-based when competitors are seat-based)

Community or ecosystem (network effects that compound over time)

Feature advantages that don't last:

"Better accuracy" (competitors catch up in one sprint)

"Faster inference" (infrastructure commoditizes)

"More integrations" (easy to copy)

The Test:

For every positioning claim, ask: Can a competitor copy this with a single product sprint? If yes, it's not defensible. Don't build your GTM on it.

Common Mistake:

Claiming you're "better" at what everyone does. In AI, benchmarks change monthly. Position on what's structurally different about your approach, not what's temporarily better about your model.

8. Ceiling Moment Qualification (Finding High-Intent AI Buyers)

The Pattern:

The highest-intent enterprise buyers for AI agents are people who've already adopted a comparable tool and hit its limits. They've invested in learning, they understand the problem space, and they have a clear business case for the upgrade.

How to Identify Ceiling Moments:

The prospect has:

Used a copilot/assistant tool for 6+ months

Hit its limitations (can't handle complex tasks, doesn't work with their stack, not autonomous enough)

Low switching costs (the mental model transfers)

Clear business case ("we're spending X hours on this manually even with the current tool")

How to Target Them:

Identify the tool(s) your AI product complements or displaces

Build target lists of companies known to use those tools

Craft outreach around the limitation, not your features:

"Teams using [incumbent] often hit a ceiling when they need [capability your product provides]"

Acknowledge the incumbent has value (don't trash-talk)

Position as "next level," not replacement

Why This Converts Better:

Ceiling-moment conversations convert 3-5x vs cold outreach because:

Prospect already understands the problem

They've already invested in the category

They have internal budget allocated

They can articulate what's missing

The Qualification Question:

"What's the most complex task you've tried to automate with your current tool, and where did it break down?"

If they have a specific answer with specific pain, they're a ceiling-moment buyer. If they say "it works fine," they're not ready.

Common Mistake:

Trying to convince tool-naive prospects to adopt AI agents. Bad conversion rates, long education cycles, and they'll compare you to "doing nothing" instead of "doing it better." Target buyers who already believe in the category.

Decision Trees

Which Positioning Should I Use?

Does your AI act autonomously (no approval per action)?
├─ Yes → Who are you selling to?
│   ├─ Developers → "Agent" framing
│   └─ Enterprises → "Teammate" framing
└─ No → "Copilot" framing

Which Pricing Model Should I Use?

Can you measure customer outcomes reliably?
├─ Yes → Outcome-based (or hybrid with outcome component)
└─ No → Continue...
    │
    Does usage vary 5x+ by customer?
    ├─ Yes → Hybrid (base + usage)
    └─ No → Seat-based

Is This Buyer Ready for AI Agents?

Do they have incident response processes for tool failures?
├─ Yes → Continue...
│   │
│   Do they have on-call rotations for production systems?
│   ├─ Yes → Qualified buyer
│   └─ No → Help them build it first
└─ No → Not ready (come back in 6 months)

Common Mistakes

1. Using "autonomous" because it sounds impressive

I've watched this slow deals. "Autonomous" scares enterprises. "Teammate" progresses faster.

2. Hiding AI failure modes

Buyers know real-world data is messy. If you don't show failures, they assume you're hiding them.

3. Treating "will it break production?" as the objection

Real objection: "who's responsible when it does?" Organizational readiness, not accuracy.

4. Pricing usage-based AI like OpenAI

Your cost structure isn't theirs. Price for 20-30% of customer's alternative cost.

5. Skipping transparency docs before demo

Order matters. Transparency → Control → Performance → Scale. Don't skip steps.

6. Demoing perfect AI

Show mistakes + recovery. Builds more trust than fake perfection.

7. Selling to buyers who demand 100% accuracy

They're not ready. Filter for mature buyers with incident response processes.

Quick Reference

Enterprise objection checklist:

 "Who gets paged when AI breaks production?" → Map to their on-call rotation

 "Who debugs AI failures?" → Map to their incident response

 "Who owns customer communication?" → Map to their escalation path

Positioning word choices:

✅ Teammate, augments, accelerates, you stay in control

❌ Autonomous, replaces, fully automated, AI-first

Demo structure:

Problem with quantified cost (30s)

AI attempt including failure/uncertainty (60s)

Human review and override (30s)

Outcome with ROI (30s)

Trust ladder:

Transparency (model card, security, explainability)

Control (approval workflows, kill switches, confidence scores)

Performance (benchmarks, case studies, live demo)

Scale (enterprise deployments, compliance, SLAs)

Pricing hybrid formula:

Base: $X/month (covers fixed costs)

Variable: $Y per unit (20-30% of customer's alternative cost)

Related Skills

positioning-strategy: General positioning frameworks and testing

technical-product-pricing: Pricing models including AI-specific patterns

enterprise-account-planning: Enterprise AI deal management

Based on enterprise AI agent GTM across developer tools and infrastructure. Patterns drawn from working enterprise deal cycles selling autonomous AI products — some carried directly, others supported alongside sales leadership — including the positioning trap diagnosis that shifted from feature competition to structural differentiation, the ceiling-moment qualification that improved outbound conversion significantly, and frameworks tested across security, operations, and engineering buyer personas. Not theory — lessons from deals where "autonomous" killed conversations and "teammate" converted.