← BUILD LOGAI-NATIVE THESIS
BUILD LOG · THE SEVENTH ANALYST

AI-native GTM is not six analysts. It is the seventh one, the one that watches the other six.

Anyone can build AI agents that do the work. The thing that makes a system compound instead of one-shot is the meta-analyst that reads the audit trails the others leave behind and turns every mistake into a training signal.

Steal this build ↓See the stackThe people part →
0.775 AUC
a won account outranks a lost one, on the model the loop rebuilt
1.5x lift
94% of the model Warm-plus accounts are wins, vs a 62% base rate
7 versions
each caught a flaw the last one shipped, by the loop not a customer
0 of 60
lost accounts that reached the top tier, the top of the funnel stays clean
THE STORY, FROM THE SEAT

Here is one I actually shipped. What happened, what I did, and the stack if you want to run it yourself.

Most AI GTM stops at automation: build an agent, let it run, move on. But an agent that never learns from its own misses just makes the same mistakes faster. The system does not get better.

It gets more confident. Six analysts doing six jobs is useful. It is not compounding.

The Sales Operator
Keep Building,
Heath
FOUNDER, THE SALES OPERATOR
THE BUILD IN FIVE MOVES · THE GIANT METHOD
  1. GMost AI GTM stops at automation, and an agent that never learns from its misses makes the same mistakes faster, more confidently.
  2. ISix analysts run on Claude Code across the Amplitude, Salesforce, and Snowflake data, each logging every decision, override, and miss.
  3. AThe analysts run and the seventh self-corrects. The operator approves each rule change before the system changes: augment, not autopilot.
  4. NThe rep's job was redrawn around judgment, and what the seventh analyst flagged became the week's coaching theme.
  5. T94% of Warm-plus accounts are wins against a 62% base rate, and 0 of 60 lost accounts reached the top tier.
RUN THE PLAY YOURSELFThis exact build is a playbook. Run it on your own data.
→
01 / HOW WE APPROACHED THE PROBLEM

Ran the loop.

Six agents doing six jobs across the product and CRM data was useful, not compounding: six jobs done well still make the same mistakes twice unless something reads the misses. So every analyst got an audit trail, and a seventh was built to read the other six's trails and turn each miss into a training signal. The exhaust of the system became its memory.

  1. 1
    Give every analyst an audit trail

    Each agent logs what it decided and why, every override, every missing field, every enrichment miss. The exhaust of the system becomes its memory.

  2. 2
    Build the seventh analyst to read them

    A meta-analyst reads the other six's audit trails and turns each mistake into a training signal: a rule tightened, a gap flagged, a pattern learned.

  3. 3
    Close the loop

    The corrections feed back into the system, so the next run is better than the last. That is the difference between a tool and a system that improves itself.

02 / WHAT IT TOOK CROSS-FUNCTIONALLY

Nobody ships this alone. Here is who had to move.

An analyst agent is only as good as the sources it reads and the people who act on what it finds. Both sat outside the build.

REVOPS AND DATA
The sources the analysts read.

The agents ran across CRM, product, and revenue data that other teams kept clean and connected. The meta-agent could learn from its misses only because the underlying sources held still.

FINANCE AND LEADERSHIP
Consuming the output.

The reports fed the same operating rhythm leadership ran on, so an agent finding got treated like an analyst finding: read, challenged, and acted on.

03 / HOW WE TURNED IT INTO A SALES MOTION · THE PEOPLE PART

The seventh analyst watches the machine. A leader still has to build the team around it.

An AI-native GTM system does not run itself, and it does not manage the humans who run it. The leadership half was defining what the reps do once the agents take the busywork, and coaching the judgment a machine cannot.

THE FRAMEWORK BEHIND IT
The Split, applied to a team

Name what the agents own, research, prep, scoring, the watch, and what the humans own, the conversation, the judgment call, the relationship. Enable reps on the half only they can do, instead of asking them to compete with the machine on the half it already won.

  1. 01
    Redrew the rep's job around judgment

    When agents took the prep, the rep's day became the conversations and the calls, and I coached to that, not to activity.

  2. 02
    Made the agents' misses a coaching input

    What the seventh analyst flagged became the week's coaching theme, not just another dashboard nobody opened.

  3. 03
    Taught the team to orchestrate the agents, not fear them

    Reps learned to run the system, so AI was leverage in the workflow, not a threat to the seat.

Build the team enablement →
04 / WHAT I LEARNED

Not proof. Just what the build taught me.

  1. 01
    An agent that never learns gets more confident

    Automation without a feedback loop does not get better, it fails faster. The misses have to feed the next run.

  2. 02
    The exhaust is the memory

    Every override, missing field, and enrichment miss got logged, then read. A silent failure became a training signal.

  3. 03
    The operator approves before the system mutates

    Corrections tighten rules and flag gaps, but a human signs each change before it ships. Augment, not autopilot.

  4. 04
    Enable reps on the half only they can do

    When agents took the prep, the rep's day became the conversations and the judgment calls, and the coaching followed that, not activity.

05 / THE WORKFLOW

The runnable version. Copy it into your stack.

Claude CodeSnowflakeDeeplineAmplitude + SalesforceThe audit trailCorrections fed back
[ SALES OPERATOR ]

The 7th Analyst

Heath Barnett · Sales Operator
G / Groundthe problem that started all of it, with the receipt and the cost
AI-native

An agent that never learns just fails faster.

Automation without a feedback loop does not get better. It gets more confident.

Drag

Six analysts is useful, not compounding.

Six jobs done well still make the same mistakes twice unless something reads the misses.

A / Assignthe build, block by block, and who owns each one. Open a step to see it run.
HUMAN + AI, IN THE LOOP
The analysts run and the seventh self-corrects; the operator approves each change before the system changes.
Run the analysts
Claude Code01
Run the six analysts
Details
Six agents doing six jobs across the product and CRM data.
AI
Snowflake02
Read and write the warehouse
Details
The warehouse the analysts read and the audit log writes back to.
AI
Audit every decision
Claude Code03
Log what each agent decided and why
Details
Every override, every missing field, every enrichment miss. The exhaust becomes memory.
AI
Claude Code04
Build the seventh analyst to read them
Details
A meta-analyst reads the other six's trails and turns each miss into a training signal.
AI
Close the loop
Deepline05
Feed the corrections back
Details
The rule tightens, the gap gets flagged, so the next run is better than the last.
AI + human
The operator06
Approve the change
Details
The operator approves each rule change before the system mutates itself. Augment, not autopilot.
Human
N / Normalizethe motion that made it stick
A workflow without a motion is dead. This is how it became the way the team works.
1

The rep's job redrawn around judgment

Agents took the prep, so the rep's day became the conversations and the calls, and coaching aimed there, not at activity.

2

Misses became the coaching theme

What the seventh analyst flagged set the week's coaching, not just another dashboard nobody opened.

3

Reps learned to run the agents

The team was taught to orchestrate the system, so AI was leverage in the workflow, not a threat to the seat.

T / Tie backthe result it drove, and how they know
RESULT · 01

16.3-point score gap

the rebuilt model opens, won vs lost.

RESULT · 02

82 claims recomputed

against source, 0 open flags.

RESULT · 03

100% top-tier precision

16 wins, 0 losses reached Hot or Strike.

OUTPUT · 01

A per-agent audit trail

Every decision, override, and miss.

OUTPUT · 02

The seventh analyst

Turns misses into training signals.

OUTPUT · 03

A closed feedback loop

Each run better than the last.

The Sales Operator
06 / HOW YOU DO IT TOO

Here is what I built. Here is how you build it.

The whole thing installs as one plugin. Or grab the skills a la carte. Everything on this page, runnable, on what you paste today.

✓ EVAL PASS · 4/4
THE PLAYBOOK · ONE INSTALL
Build the AI-Native GTM Operating System

11 skills chained into one runnable play. Installs as a single plugin, no copy-pasting each skill. It runs on what you paste; connect your stack to go live.

SEE THE FULL PLAYBOOK →
SEE IT RUN IN CLAUDE
EXAMPLE CHATDev, VP GTM, running the loop in one sitting
D
Everyone tells me to hire a GTM engineer. Before I add headcount, where do I actually start building the AI-native engine myself?
G
Ground· Solve the problem
Do not automate the org chart. Pick the one workflow that costs you the most this quarter and frame it as a workflow, not a hire.
Receipt
The crux: reps hand-prep every account. About 9 hrs a week each, 40% on accounts that never convert. Fix that workflow, not the headcount.
I
Identify· Stack the tech
Give that workflow the right context. Buy the plumbing, build the thin edge, and name the ICP tiers this workflow runs on.
Receipt
A stacked workflow: fit plus timing signals, enrichment, and the ICP tiers this workflow qualifies against. Bought the plumbing, built the edge.
A
Assign· Cut the drag
Split the non-selling drag from the judgment. Scoring, enrichment, and first-draft prep come off the rep plate, and the calls that need a human, who to pursue, the message, stay named, owned, and reviewed weekly.
Receipt
About 7 hrs a week back to selling per rep. The system preps; the rep decides and sells.
C
Close the loop· Pattern analyst
Then wire one signal to one booked motion. Pattern Analyst feeds outcomes back into fit, QA Agent watches for scoring and CRM drift, and Evolution Agent reads those findings and proposes the next tweak for you to approve. That is what makes it compound instead of a one-off.
Receipt · the loop closed
One workflow framed instead of hired, a context pack, a signal wired to a motion, and a loop that learns. The GTM-engineer req never went out.
THE SKILLS IT USES
Find what is actually broken in the workflow.
solve-the-problem
Take the non-selling work off your reps.
cut-the-drag
Keep reps making the calls that matter.
keep-the-judgment
THE OTHER HALF · LEAD THE TEAM
Build the team enablement→
I write up one of these a week. Free, receipts only.
RELATED BUILDS
The Win/Loss Backtest
The #1 predictor of a win was not in the model
The Governance File
It failed three ways in a week. Then it couldn't.
THE MARKET ON THIS
GTM 3.0: humans, agents, and multi-agent workflows (GTM Strategist)Who is an AI GTM engineer? the revenue systems architect (Apollo)