The Constraint Question — ETJ Life CEO Operating System Series, Vol. 02

CEO Operating System Series · Volume 02

Stop asking which AI tool. Start asking a better question.

A short, scannable Field Note. Click any header or card to go deeper. Companion to the AI Efficiency Playbook for PE-Backed CEOs.

Built on pattern recognition from

212 live CEO
conversations across
146 PE-backed companies

→

How to read this: Scan the section titles. Click any card or header to expand. End with the diagnostic to produce a one-page brief.

“

The wrong question is "what AI should we buy?"
The right one is "what's actually in our way?"

— The Operating Principle

§ 01 / The Wrong Way In

Why most leaders burn 18 months on AI before any return.

Tool first, problem second, business model never. The leverage is upstream.

The Question Most Ask

"What AI tool should we try first?"

The Question That Works

"What constraint, if removed, changes the economics of how we work?"

Walk into any leadership team conversation about AI right now and you'll hear the same anxious music. Where do we start? Which platform is the right one? Are we behind? What if we waste the budget? What if a customer sees something hallucinated and we end up in a screenshot?

Reasonable fears. Wrong starting point. Most CEOs and operators are approaching AI the way a 1995 board approached the internet — tool first, problem second, business model never. They evaluate platforms before they've named what they're trying to change. They subscribe before they've defined a constraint. They count seats instead of outcomes.

Twelve to eighteen months later, the seat licenses have renewed, the consultancies have been paid, and nothing in the P&L has actually moved. The shift isn't getting smarter on tools — tool selection is the smallest, most reversible part of this. The leverage is upstream, in the question you ask before you open a vendor list.

§ 02 / The Economic Frame

Three flows. Know which one you're standing in.

Most AI plans are 90% Flow 01. That's why most AI plans don't move enterprise value. Click each flow to see the trap.

FLOW 01

Cost extraction — same work, fewer hours.

Real, but commoditizing. Every competitor will get here within 24 months, and the savings get repriced into market expectations. This funds the program. It doesn't differentiate it. Necessary, not sufficient.

FLOW 02

Capacity creation — do work that previously broke the economics.

Serve a tier-2 segment, run personalized outreach at scale, ship faster, qualify deeper. Margin-accretive and harder to copy — because it requires workflow change, not just tool adoption. Tools are commoditized; how you use them isn't.

FLOW 03

Category repositioning — AI changes what the business is.

New product, new pricing model, new segment, new moat. Rare. Hard. The biggest unlock and the only one that survives sponsor turnover. This is the one your PE board actually cares about — though they may not say it that way. Flow 01 funds the experiment. Flow 02 funds the next round. Flow 03 funds the exit.

§ 03 / The Framework

Four constraints. That's the whole map.

Almost every business bottleneck reduces to one of four barriers. Click each letter.

Data

"We don't have it. Or can't reach it."

Scale

"We can't reach enough, fast enough."

Time

"There aren't enough hours."

Capability

"No one can actually do this."

// 2026 SHIFT

From "AI tools" to agentic workflows.

Until recently, "using AI" meant a human prompted a chatbot. That era is closing. The new unit of leverage is the agent — software that runs a multi-step process end-to-end on its own cadence, with human approval gates at the right moments.

What that means: the cost of running a process is decoupling from the cost of staffing it. Constraints that used to live in T (Time) and C (Capability) collapse — but two adoption pressures show up harder: trust in the output, and workflow integration. When an agent does the work, "is the output good?" matters more than "did we save hours?" — and "does it fit how decisions get made?" matters more than "is the model state of the art?"

§ 04 / The Protocol

Four steps from constraint to commit.

Each step has an owner. Hover or tap to see who.

Name & size.

One bottleneck. Specific, bounded, repeated. Then quantify the prize: dollars, hours, deals, NPS. The CEO who can't size the prize won't get the budget.

Owner: Function Lead + CFO

Dimension.

Which of D, S, T, C is most in the way? Often two. Pick the dominant one — that determines the starting move.

Owner: Function Lead + CEO

Pilot.

Choose path (Build / Buy / Borrow — see § 05). One workflow, one team, two weeks, internal-facing. Reversible. Owner with skin. Kill criteria written down.

Owner: Named Operator + CEO

Industrialize.

If the constraint moved: name the production owner, set the metric, schedule the review, put it in the L10. Without this, pilots die alone.

Owner: Function Lead + COO

§ 05 / The Decision Most CEOs Get Wrong

Build. Buy. Borrow. Choose deliberately.

The right sequence is also the right time order. Click each path.

Borrow

Rent a frontier model.

Use a horizontal LLM (Claude, GPT, Gemini) directly via chat or API. Zero infrastructure. Lowest cost to test. Every AI initiative should start here. If a borrowed model can't move the needle in two weeks of real use, the buy/build decision is premature.

Right whenYou're testing a constraint hypothesis, output is reviewed by a human, and the workflow is exploratory.

Buy

Adopt a vertical SaaS with AI inside.

Specialized tool for a specific job: Gong for calls, Harvey for legal, Cresta for support, EvenUp for claims. Faster to value than building. Captures vendor R&D spend. The trap: stacking too many vertical tools creates integration debt that erodes the savings.

Right whenThe use case is well-defined, vendor competition exists, and integration cost is materially lower than build.

Build

Ship a proprietary system on top of frontier models.

Custom workflows, agents, or applications using your own data and judgment. Only justified when the capability is competitively defensible — meaning the data, workflow, or judgment is uniquely yours. Otherwise you're paying engineering costs to rebuild what's available off the shelf.

Right whenThe data is proprietary, the workflow is differentiating, or the integration is the moat — not the model.

A useful test: "If our biggest competitor adopted the exact same tool we did, would our economics still improve?"

If yes, buy or borrow. If no — if the value depends on the specific way we do the work — consider building. The borrow-buy-build sequence is also the right time order, not just the right list. Skipping straight to build is how 70% of mid-market AI budgets get burned.

§ 06 / Apply It

The constraint diagnostic. Output: a one-page brief.

A real bottleneck, six steps, a fundable brief. Hand it to your sponsor.

Step 01 of 06 — Name it

What's one bottleneck in your business right now — one repeated process or friction point that costs you every week?

Be specific enough that a frontline operator could read it and recognize themselves in it.

Step 02 of 06 — Dimension it

Which of the four is most in the way?

Pick one, even if you see two. The dominant constraint determines the starting move.

Step 03 of 06 — Size the prize

If that barrier vanished tomorrow, what would change — and what's the economic envelope?

Be specific in dollars, hours, deals, retention, NPS, or share. The CEO who can't size the prize won't get the budget.

Step 04 of 06 — Value flow

Which flow does this value live in?

Be honest. Most initiatives are Flow 01. That's fine — but say so.

Step 05 of 06 — Owner & path

Who runs the pilot, and which path?

No accountable owner = no real initiative.

Pilot Owner

Kill Date

Build / Buy / Borrow

Step 06 of 06 — Your one-page brief

Hand this to a sponsor.

§ 07 / Traps

Four failure modes we see every week.

If your AI initiative is stuck, it's almost certainly one of these. Hover or click any trap to see the tell.

Trap 01

Tool-shopping before constraint-naming.

Evaluating platforms before naming what's in the way. The platform isn't the answer to a question you haven't asked.

The tellYour AI strategy is a vendor list, not a constraint list.

Trap 02

Pilot purgatory.

Three pilots have been "running" for nine months. Nobody owns deciding. No kill criteria. Nothing has been industrialized.

The tellYou can name the tools. You can't name the production deployment.

Trap 03

Building when you should borrow.

Six months into a custom build that a $30/seat tool already does. Engineering capacity burned on commodity capability. Or the inverse: stacking five vertical SaaS tools to recreate a workflow unique to your business, when the integration debt is now larger than the build would have been.

The tell"We're building because vendor X doesn't quite do what we need." It does. You haven't read the docs.

Trap 04

Confusing productivity with leverage.

Saving 30 minutes per person is fine. The prize is doing things you couldn't do at all — serving a tier you ignored, reaching a segment you couldn't, shipping work that previously needed a hire. Don't optimize the existing job. Reimagine it.

The tellYour AI ROI deck is in hours saved, not in revenue, segment, or capability unlocked.

§ 08 / The Handoff

Where this lives in the Playbook.

28 use cases across 10 functions. The constraint you named almost certainly lives in one of these.

Finance & FP&A

2 · Reporting, cash, working capital

Operations

4 · Procurement, inventory, freight

Sales & Revenue

3 · CRM, pricing, forecasting

Marketing

3 · Content, ABM, attribution

Customer Success

3 · Support, churn, signal

HR & People Ops

3 · Sourcing, performance, retention

IT & Infrastructure

2 · Security, helpdesk, infra

Product & Engineering

2 · Velocity, roadmap, QA

Legal & Compliance

2 · Contracts, risk, regulatory

Executive Office

4 · M&A, planning, board prep

Volume 01 · Companion Piece

The AI Efficiency Playbook for PE-Backed CEOs

Open the Playbook →

§ § §

The leaders who win this cycle won't have the most tools. They'll ask the better question.

Name and size it. Dimension it. Pilot it. Industrialize it. The tool was always the easy part.