WonderLead
Slides ← → · Notes N · Timer T
// Talk · Patricia Juárez Muñoz

Adopt AI without
losing your judgment

Two silent risks, a quality bar you can hold without slowing down, and two tools you can use today: AI Builder Stack and RePLAN.

AI slop
Cognitive risk
Builder Stack
RePLAN
WonderLead home QR

WonderLeadsite.wonderlead.tech

// About

I help people ship with AI
without losing the bar

I'm Patricia Juárez Muñoz: Senior Tech Lead, AWS EQ Champion, founder of WonderLead. I build practice fields for AI-native work: quality standards, intentional stacks, and planning systems people actually follow.

WonderLead OS
Practitioner · Builder Stack · RePLAN · Community
Today's job
A mental model for adoption that protects judgment, plus two tools you can use immediately.
// Agenda

Where we are going

1
The speed gap: adoption faster than critical thinkingName the problem in plain language
2
Two silent risks: AI slop + quiet judgment lossEvidence, patterns, the rubber-stamp review
3
What quality requires when AI does part of the workSpec · generate · evaluate · hold the bar
4
Live: Builder Stack, then RePLANAnti-FOMO stack choice + intentional skill roadmap
// The speed gap

AI adoption is moving faster
than our ability to think critically about it

Tools arrive weekly. Standards do not. Most teams treat "use AI" as the strategy. That is not a strategy. That is velocity without a bar.

More output
Less shared judgment
No playbook
// Risk 1

AI slop: polished, thoughtless

Slop is low-quality AI output that looks competent. It compiles. It often passes tests. It is still expensive to review and dangerous to maintain.

01
Superficial competenceLooks right until load, edge cases, or change
02
Asymmetry of effortTrivial to produce, costly to remediate
03
Mass producibilityVolume scales. Quality does not.
// Slop patterns

What slop looks like in practice

Naming
Generic names: data, result, handler. Code that cannot be searched or owned.
Hollow docs
"Seamlessly integrates." "Ensures quality." Marketing language with zero density.
Architecture drift
New patterns for problems the codebase already solves. Silent divergence.
Hallucinated APIs
Fake packages, invent methods, happy-path security gaps.
// Evidence · keep light

The data is not subtle

1.7x
More issues in AI co-authored PRsCodeRabbit · logic and correctness errors ~75% higher
4–8x
Growth in duplicated code blocksGitClear · refactoring activity collapsed as AI usage grew
Net
Quality gains offset by stability dropsVelocity without durability is a trap
2x
Higher revert / rewrite rates for AI-assisted codeWhat ships fast often comes back as debt
// Risk 2

The quiet risk: judgment atrophy

Cognitive offloading is useful when it frees you for harder thinking. It is harmful when it replaces thinking. GPS does not make you a better navigator. Uncritical AI use does not make you a better engineer or PM.

Higher confidence in AI → less critical thinking.
Microsoft / Carnegie Mellon pattern
// Atrophy cycle

Three phases

Phase 1
DependencyYou delegate harder problems because the model "works" for task completion
Phase 2
AtrophyIndependent decomposition weakens from disuse. Confidence in your own judgment drops.
Phase 3
Bias propagationYou internalize AI idioms as correct. You pass them on in review and mentorship.
// Review crisis

Rubber stamps are the real outage

Automation bias: we over-weight machine output and under-weight independent scrutiny. The illusion of oversight is worse than no oversight.

More to review
AI creates review burden: more issues, more context required.
Less scrutiny
Same reviewers are less likely to exercise that burden.
Fix
Require one risk surfaced or one change made. "LGTM, no notes" is a signal.
// What quality means now

Five dimensions when AI helps

Factual accuracy
Claims grounded in context or verified sources
Hallucination
Zero tolerance for invented APIs in code
Consistency
Agrees with itself and system docs
Format
Matches schema, style, standards
Architecture
Follows established patterns, not drift
// Operating loop

Agents execute. You think.
The bar holds.

01

Agents execute

Draft, scaffold, search, generate boilerplate. Speed where speed is cheap.

02

You think

Problem framing, trade-offs, edge cases, customer impact. Judgment stays human.

03

Bar holds

Spec, eval, review. No merge without ownership of what shipped.

// Spec before generate

If you cannot specify it,
you cannot evaluate it

1
Write the interface, inputs, outputs, and edge cases first
2
Then prompt. Not the other way around.
3
Attach the intent to the PR so reviewers ground themselves
4
Happy path is free. Edge cases are where slop hides.
// Fresh eyes

Generator ≠ Evaluator

Having the same context grade its own work is like an author proofreading their draft in one sitting. Use a fresh eval pass: different session, checklist, or person.

Planner
Brief → Spec
Fill gaps before generation
Generator
Scoped sprints
One feature, one commit
Evaluator
Fresh context
Grade against the spec, not the vibe
// Hold the bar without slowing

Structured offloading

AI does
Boilerplate, format conversion, first drafts, doc search, test data
You do
Design decisions, edge cases, architecture, customer impact, trade-offs
Ritual
After every AI pass: "What did I verify? What did I correct?" Two sentences.

Speed comes from clear ownership of judgment, not from skipping verification.

// Mechanisms

"Good intentions never work. You need good mechanisms to make anything happen."

Personal mechanisms: spec-before-prompt, fresh eval, verification ritual, intentional stack, intentional practice roadmap. Intentions do not survive FOMO week.

// Bridge

From FOMO to intention

A
Tool chasingEvery release, every thread, no chosen stack
B
Reactive learningWeekend experiments that never become habits
→
Two mechanisms nextBuilder Stack: choose. RePLAN: practice on purpose.
Builder Stack QR

Builder Stackai.wonderlead.tech

// Live demo 1

AI Builder Stack

Choose tools matched to your role and stack, instead of chasing every release out of FOMO.

1
State role, work type, constraints out loud
2
Land a recommended stack you can defend
3
One tool deep: why it fits, not a feature list
// Demo 1 · takeaway

You leave with a chosen stack

Not a longer tool list. A decision you can explain. That is the anti-FOMO mechanism.

Builder Stack
Find your stackhttps://ai.wonderlead.tech/builder-stack/
RePLAN QR

RePLANreplan.wonderlead.tech

// Live demo 2

RePLAN

Turn AI skill-building from reactive to intentional. Produce a personal roadmap you actually follow.

1
Reflect: name the real friction
2
Prioritize 1–2 skills for 4–6 weeks
3
Put them on a weekly cadence
4
Show the roadmap, not a wishlist
// Demo 2 · sample roadmap

Example 6-week practice

Wk 1–2
Spec before generateWrite edges before every non-trivial prompt
Wk 3–4
Fresh eval checklistSeparate generate session from evaluate session
Wk 5–6
Stack lock-inBuilder Stack choice + stop tool FOMO
// Takeaway

Leave with this

2 risks
AI slop + quiet judgment loss
1 loop
Agents execute · You think · Bar holds
2 tools
Builder Stack + RePLAN
1 ritual
What did I verify? What did I correct?
// Start today

Use them in the community

Scan, choose your stack, build your roadmap. Quality is a habit, not a talk slide.

Builder Stack
Builder Stack QR

Find your stack

Role-matched tools, not FOMO.

Open
RePLAN
RePLAN QR

Build your roadmap

Intentional AI practice.

Open
// Stay in the loop

Newsletter + deeper practice

Free for new subscribers: How to Ship with AI handbook. Paid gift: AI Quality & Cognitive Guardrails handbook.

Newsletter
Subscribewonderlead.tech/subscribe
Community
Practice with people who hold the bar with you.
Optional next
Practitioner: team practice under Working Backwards reviews.
Notes
Title

Open with the thesis. Do not sell tools yet.

Do Ask who adopted a new AI tool this month without a written quality bar.

Notes
About

Keep short. Point to QR home.

Notes
Agenda

Promise demos late so people stay for the mental model first.

Notes
Speed gap

One story of a polished PR that still failed review.

Notes
Slop def

Slop is not "AI is bad." Slop is unowned AI output.

Notes
Patterns

Pick one pattern the room laughs at. Move on.

Notes
Evidence

Do not read every number. Hit 1.7x and duplication.

Notes
Cognitive

GPS analogy. Pause. Let it land.

Notes
Cycle

Phase 3 is the organizational risk: mentorship spreads atrophy.

Notes
Rubber stamp

Require one risk or one change. Mechanism, not guilt.

Notes
Dimensions

Architecture coherence is the sleeper: drift across many small PRs.

Notes
Loop

Repeat the line twice. It is the talk brand.

Notes
Spec

Live mini: write 3 edge cases before any prompt.

Notes
Eval

Fresh context = new chat or checklist or teammate.

Notes
Offload

Teach the two-sentence ritual. Have them write it once.

Notes
Mechanisms

Bridge to tools: mechanisms you can open in a browser.

Notes
Bridge

Transition: demos start now. Reset Builder Stack tab.

Notes
Stack demo

Script: FOMO 30s → intent 30s → live find stack 3–4m → defend one tool → close.

Fallback Screenshot if network fails.

Notes
Stack takeaway

Chosen stack vs longer list. QR dwell 10s.

Notes
RePLAN demo

Script: reactive learning 45s → Reflect → Prioritize 1–2 skills → weekly cadence → show roadmap.

Notes
Sample roadmap

Fallback if RePLAN login fails. Still narrate habits.

Notes
Takeaway

Four cards. Ask audience to screenshot.

Notes
CTAs

Primary: Stack + RePLAN. Soft: newsletter. Practitioner optional one line.

Notes
Close

Mention free vs paid handbooks. Thank. Open Q&A.

1 / 24