Chicago Booth School of Business  ·  Chief AI Officer Program

AI-Native
Development

What every Chief AI Officer needs to know —
and what to do about it this quarter
Richard Cross  ·  Enlyt Building Technologies
richard.cross@enlyt.io
The premise

One question changed everything.

Before
Can we
build this?
Gated by developer time, cost, and availability. The answer was usually no or not yet.
→
Now
Should we
build this?
The bottleneck is now judgment, not code. The answer is almost always yes, try it.
Coding time has dropped 10–100×. Prototyping is nearly free. That doesn't change the technology — it changes every organizational decision downstream of it.
The data

Where the time goes has shifted dramatically.

2023
Planning & Requirements15%
Development60%
Testing20%
Deploy5%
2026
Planning & Requirements22%
Development12%
Testing & Judgment54%
Deploy12%

Development shrank from 60% → 12% of total effort. Requirements and testing together now account for 76%. Your organizational investments must follow the work — not the old model.

Discussion
Where is your organization's SDLC time currently allocated?
What percentage of your software spend goes to raw development vs. requirements and testing?
When a project fails, is it usually a development failure — or a requirements failure?
How much of your testing is done by engineers vs. domain experts and end users?
Shift left

Requirements clarity is the new bottleneck.

← Shift Left
Invest more in Requirements
AI amplifies ambiguity. A vague specification produces a vague product — built 10× faster. Every hour of requirements clarity saves 10 hours of rework.
What to hire / promote Product managers, business analysts, and domain experts who can write precise, testable requirements. They are now your most leveraged people.
→ Shift Right
Invest more in Judgment
AI builds fast. Evaluating what it built is the new critical path. Acceptance testing — "did it build the right thing?" — is now the job only humans can do well.
What to build / fund Test infrastructure, user research, domain expert review cycles. QA is no longer a cost center — it is the primary value delivery mechanism.

The implication for headcount: You need fewer junior developers and more senior "directors" who write clear specs and judge quality output. The 10× developer of 2026 writes clear requirements, not more lines of code.

Discussion
In your organization, who owns requirements quality?
Do you have a process for writing requirements that are clear enough for AI to implement? Who owns it?
Who currently does acceptance testing — engineers or domain experts? Should that change?
What would it mean to treat requirements writing as a core organizational competency rather than a step in a process?
Strategic decision

The build-vs-buy calculation has fundamentally changed.

Dimension Buy (SaaS) Build — Traditional Build — AI-Native
Build cost$0$300K – $1M$20K – $100K
Time to valueDays6 – 18 months2 – 8 weeks
Feature fit~80% — vendor roadmap decides the rest100% — you control it100% — you control it
Annual cost$30K – $150K/yr (and rising)$80K – $200K/yr maintenance$20K – $50K/yr maintenance
Competitive advantageNone — competitors have the same toolHighHigh
Vendor lock-inHighNoneNone
Old conclusion
Build cost was 5–10× SaaS cost. Buy, unless the need is truly unique.
New conclusion
Build cost is now roughly equal to or less than SaaS — with 100% fit and no lock-in. Default to build unless speed-to-market is critical.
Evidence

Three organizations that changed the calculation.

Enlyt / Construction
Bid Automation
Each construction bid required a day: reading specs, calculating materials, writing narrative, formatting. Replaced with a Claude workflow.
1–2 hrs
per bid (was a full day) · 338 projects automated
DonkeyVan / Transport
Custom Dispatch App
Client needed a working Uber-style dispatch demo before committing to development. Expensive dev shop engagement quoted. Built with AI instead.
Hours
to build full presales demo + pitch deck
Minesweeper.org
Replacing the Stack
6 SaaS products replaced: Jira, GitLab CI, DataDog, WordPress CMS, translation agency, SEO consultant — all replaced by Claude-driven workflows.
$0
incremental SaaS spend for workflow tools

The pattern: In each case, the SaaS or agency alternative would have cost $30K–$150K/year and delivered generic capability. The AI-native custom solution delivered better fit, faster, at a fraction of the cost.

Discussion
Which SaaS products in your stack could you build and own instead?
List three tools your organization pays for that have low feature fit — you only use 60–70% of what they offer.
For each: what is the annual cost? What would a custom AI-native build actually require?
Which workflows do you currently outsource (agencies, consultants) that are pure pattern-matching — and could be automated?
Organizational action

Reallocate toward judgment, away from production.

Reduce investment in
Junior developer headcountWriting code is no longer the bottleneck. Reduce hiring of entry-level coders.
Generic SaaS subscriptionsReview the stack annually. Build what you can own.
Manual, repetitive documentationBids, reports, status updates, proposals — automate the pattern.
→
Increase investment in
Senior "directors"People who write clear specs and judge quality output. The 10× developer now sets direction, not lines.
Domain expert testersAcceptance testing by people who know the domain is the new critical path.
Requirements engineeringProduct managers, BAs, user researchers. Clarity before build is now your highest-leverage investment.
AI infrastructure & toolingClaude Code, model API access, prompt management, evaluation frameworks.
Action

Three entry points. Pick one this week.

01
For the C-Suite
Automate a recurring document

Identify one document your team produces every week — a bid, a report, a status update, a proposal. Open claude.ai, paste your template, ask Claude to fill it. Try it this week. Measure the time saved.

No code, no IT. One afternoon.
02
For internal operations
Build one internal tool

Give Claude a dataset and ask for a web dashboard or decision support tool. Start with something internal where mistakes are safe — a competitive analysis tracker, a proposal scoring tool, an operations report.

Low risk. High signal on organizational readiness.
03
For product leaders
Run one build-vs-buy experiment

Pick one SaaS product you're about to renew. Get a quote to build it AI-native. Compare feature fit, cost, and timeline. Even if you choose to renew, you now have real data for the next decision.

The exercise changes how you see every future procurement.
Summary

Three things to take back to your organization.

Shift Left
Requirements clarity is your highest-leverage investment

The cost of ambiguity is 10× higher when AI can build fast. Fund requirements engineering, user research, and product management. Promote the people who write clear specs.

Shift Right
Judgment and acceptance testing are the new critical path

Testing shrank when developers wrote the code. It expands when AI does. Your domain experts, QA leads, and user panels are now central to delivery — not a final checkpoint.

Recalculate
Default to build — buy only when speed matters most

The build-vs-buy math has changed. Review your SaaS stack. Audit your agencies. Any workflow that is pure pattern-matching is a candidate for replacement. Start with the highest-cost line item.

Chicago Booth  ·  Chief AI Officer Program

Questions?

richard.cross@enlyt.io minesweeper.org/vibecoding github.com/ajaxchess/minesweeper.org