Dharmesh Shah’s documented lens
Broad can work when integration burden—not feature depth—is the customer’s core problem. AI synthesis, not a direct quote.
Source · Apr 2024 ↗Judgment, accelerated01 / 15
An executive walkthrough
How I use AI to make better product decisions.
AI sharpens judgment at three moments. I still own every decision.
Before I commit
Ideate
Challenge the assumption.
Before I align
Align
Make the future tangible.
While we deliver
Deliver
Reconcile signal with evidence.
Guardrail: invented, personal working examples. Production use belongs inside approved tools, approved sources, and established controls.
The operating habit
Three moments. One decision thread.
Before I commitIdeate
Broad bet → conditional decision
Before I alignAlign
Strategy memo → clickable experience
While we deliverDeliver
Ticket noise → evidence-backed signal
AI handles interrogation and synthesis. I own the evidence, the decision, and the consequences.
01 · Ideate — Napkin → Narrative
Start with the conviction.
Then find the fracture.
Am I making the right bet?
FieldFlow · flawed conviction
“Distribution is the whole game. Integrate broadly now; optimize attach later.”
11%attach vs. a 30% planning assumption
Human providesA conviction, context, and the decision at stake
AI uniquely doesInterrogates weak assumptions and retrieves opposing evidence
Human decidesWhat changes—and what test settles the argument
Ideate · tension before confidence
Two questions changed the work.
1What evidence distinguishes a discovery problem from a workflow problem?
2What has to be true for breadth itself to solve the customer’s problem?
Jake Fuentes + Annie Pearl lenses
Breadth across disconnected buyers frays the ICP; focus is a choice about where to play. AI synthesis, not a direct quote.
Sources · 2023 ↗Derived analysis from a frozen Lenny’s Newsletter / Podcast subscriber-archive snapshot. Source credit: Lenny Rachitsky. Not affiliated with or endorsed by Lenny’s Newsletter.
Ideate · from slogan to test
The output is not better prose.
It is a better decision.
Go broad now. Fix attach later.
↓
For six weeks, isolate the cause of 11% attach in field services. Go deep on two verticals if workflow gaps dominate; stay broad only if discovery is the primary issue and one shared integration materially fixes it.
TimeboxSix weeks
BranchWorkflow gap → go deep
ExceptionDiscovery problem + shared fix → stay broad
01A · The custom analysis engine
A prompt can draft.
A tool can remember how to think.
624posts + episodes
15,669typed evidence fragments
4,367opposing-view pairs
345guest profiles
01Source archiveA dated, licensed personal snapshot of posts and transcripts
02AI extractionFrameworks, claims, contexts, caveats, and contrarian positions
03Tension graphA second pass connects positions that genuinely disagree
04Layered memoryFast summaries, curated evidence, and source-level escape hatches
Frozen personal archive snapshot; no raw subscriber-archive files are redistributed. Source credit: Lenny Rachitsky. Not affiliated with or endorsed by Lenny’s Newsletter.
The engine at decision time
Bring in only the evidence this claim needs.
Read the claimFind what must be true
Retrieve by meaningNot just keyword match
Prefer tensionDefender + critic + condition
Return a traceSource, date, status, human check
Many audiences, one use case, is not a coherent horizontal product.
Jake Fuentes argues horizontal products work when one audience has many connected needs.
Newsletter · Oct 31, 2023 ↗Breadth works when integration is itself the customer problem.
Dharmesh Shah’s HubSpot example supports broad capability when fragmented tools create the burden.
Podcast · Apr 4, 2024 ↗The model proposes; the evidence constrains; I decide.
02 · Align — The Playable Roadmap
Make the strategy tangible enough to react to.
Can everyone see the same future?
Human providesThe chosen strategy, north star, and constraints
AI uniquely doesTranslates each stage across experience, platform, value, and controls
Human decidesWhat is coherent, feasible, and worth funding
Strategy and product evolve together—so the room debates one shared future, not six interpretations of a memo.
CrawlProve the workflow
WalkMake it repeatable
RunManage the money
Align · Crawl — prove the workflow
Turn an accepted estimate into a deposit request.
Without making the contractor switch tools.
ExperienceNorthside converts a signed job estimate into one secure deposit request.
ValueProve workflow placement lifts attach before scaling distribution.
ProofDeposit attach · time from estimate to deposit
ControlsExplicit consent; identity checks; funds settle to the existing business account.
New capabilities: Payment link · Merchant onboarding · Settlement reporting
Northside HVACCrawl · illustrative
Job 2407 · accepted
Ridgeway boiler replacement
Estimate$8,400
Deposit due$2,100
Request secure deposit
Align · Walk — make it repeatable
Make payment part of the job.
Not a separate collections task.
ExperienceApproved milestones trigger a payment request; authorized methods can be reused.
ValueFewer manual touches and more predictable job-level cash flow.
ProofMilestone attach · manual touches per job
ControlsClear authorization and revocation; auditable trigger history; dispute handling.
New capabilities: Token vault · Job-event triggers · Dispute workflow
Northside HVACWalk · illustrative
Job 2407 · milestone 2
Equipment installed
Milestone approvedToday, 2:14 PM
Payment request$3,150
Send milestone request
Align · Run — manage the money
Make job cash explainable.
Inside the software that runs the work.
ExperienceNorthside sees expected inflows, obligations, and reconciliation by job.
ValueLess spreadsheet work and earlier visibility into cash gaps.
ProofReconciliation time · forecast variance
ControlsRole-based access; forecasts explain inputs; no autonomous movement of money.
New capabilities: Cash ledger · Role permissions · Reconciliation rules
Northside HVACRun · illustrative
Next 14 days · explainable view
Job cash outlook
Expected inflows$28,900
Committed costs$19,400
Projected cushion$9,500
Show assumptions
03 · Deliver — Team Pulse
Reconcile how work feels with how work moved.
Can the organization deliver the bet?
What the team raised
Ranked aggregate themes from scores plus free text.
01Partner sandbox stability8 mentions
02Review queue load6 mentions
03Webhook retries5 mentions
04Integration ownership3 mentions
What the work record grouped
AI-proposed themes, each reviewed against linked tickets.
01Partner sandbox38 tickets · 4.8d
02Integration ownership22 tickets · 6.2d
03Webhook retries17 tickets · 3.1d
04Review queue14 tickets · 2.7d
Human providesAnonymous pulse, scoped work record, and decision context
AI uniquely doesGroups prose, links evidence, and compares the signals
Human decidesWhich hypothesis to investigate and what system to change
Deliver · the gap between the signals
A normalized boundary is slowing the roadmap.
Team Atlas · 90-day review
Integration ownership
Only three pulse mentions—but 22 tickets and the longest median added wait.
ATL-184“Waiting for platform to confirm which team owns the merchant-state mapping.”
ATL-207“Moved back to Atlas after two days; integration team says the adapter is ours.”
ATL-231“Blocked again on the same ownership question from the milestone-payment flow.”
Directional method: added wait = first blocked transition to first subsequent active transition. Median shown; overlapping themes are not summed.
AI proposes themes. A human reviews every theme before leadership sees it.Read-only · Scoped fields · Names redacted · No individual scoring
Invented example using 412 synthetic tickets and 9/11 fictional pulse responses. No customer, employee, or Chase data.
The operating habit
Three moments.
One standard.
Ideate
Challenge the assumption before commitment. Turn conviction into a falsifiable decision.
Align
Make strategy tangible before consensus. Let the room react to one shared future.
Deliver
Reconcile signals before intervention. Act on evidence, not the loudest anecdote.
AI handles interrogation and synthesis. I own the evidence, the decision, and the consequences.
A responsible next step
Run one bounded pilot.
Important enough to learn from. Safe enough to stop.
Start with one decision.
Use an approved environment and sanitized or synthetic inputs.
One named decision owner and one human reviewer
Read-only, minimum-necessary source access
Success, stop, and escalation criteria agreed in advance
Every important claim traceable to evidence