Bottom-Up Processing in a Top-Down AI World: Why Neurodiversity Is the Ultimate HITL Engine

For over twenty years, my world has been governed by gates, pipelines, and the relentless quest to catch broken logic before it ships. But as generative AI and agentic testing sweep through our delivery architectures, the nature of "testing" itself is undergoing a fundamental paradigm shift.

We are moving away from manual script execution and toward Human-in-the-Loop (HITL) oversight. Yet most discussions around HITL focus solely on where to place the human in the pipeline, rather than what kind of human cognition needs to be standing there.

If we want to build resilient Quality Intelligence in an AI-driven era, we have to talk about cognitive diversity—specifically, why neurodivergent bottom-up processing is the ultimate safeguard against top-down algorithmic blind spots.

The Architecture: Top-Down vs. Bottom-Up

To understand why neurodiversity is vital to modern QA leadership, you first have to look at how Large Language Models (LLMs) construct reality:

• LLMs operate Top-Down: AI models predict output based on historical statistical averages. They synthesize vast amounts of training data, identify probabilistic patterns, and generate a plausible, smooth result. They excel at generalities, boilerplate, and the "happy path."

• The ADHD Brain operates Bottom-Up: A hot-running, neurodivergent brain doesn't start with the smooth average. It starts with granular details, edge cases, anomalous inputs, and systemic friction points. It processes data from the ground up, noticing the single misplaced variable that invalidates the broader pattern.

In traditional waterfall bullpens and legacy corporate environments, a bottom-up, high-speed ADHD engine was often mislabeled as a "system variance"—an unpredictable brain that struggled with repetitive manual loops or rigid, unoptimized administrative processes.

In the GenAI era, that exact cognitive architecture is no longer a variance. It is an enterprise asset.

Why "Happy Path" AI Needs "Cynical Scrutiny"

AI engines can generate base-layer test coverage in seconds—slashing manual authoring bottlenecks from hundreds of hours down to manageable frames. But because AI predicts outputs based on prior probabilities, it suffers from a natural, structural limitation: it struggles to anticipate unprecedented failure modes.

When an LLM outputs code, test cases, or clinical data summaries, it looks convincingly correct. It feels right because it was trained to simulate success.

This is where the bottom-up human tester steps into the loop:

1. Edge-Case Sensitivity: Where an AI sees a continuous line, a bottom-up processor sees the micro-fractures in stored procedures, database schemas, and API payloads.

2. Contextual Risk Evaluation: An algorithm optimizes for raw execution speed; a human evaluates the social contract of the code—asking whether a feature safeguard protects end-user trust and clinical compliance.

3. Disrupting the Average: High-performing QA isn't about validating that the system works when everything goes right; it’s about having the intuitive curiosity to push the system into the exact state where its assumptions collapse.

Defragmenting the Loop: What Leadership Must Do

If Quality Intelligence is about pairing algorithmic efficiency with human empathy and critical scrutiny, then leadership must evolve beyond simple capacity metrics.

To truly leverage Humans-in-the-Loop, engineering leaders must:

• Protect Capacity for High-Value Oversight: Offload the repetitive 80% of boilerplate authoring to AI, freeing senior engineers to focus exclusively on deep-system validation and exploratory risk analysis.

• Embrace Cognitive Diversity: Stop trying to force every tester into a rigid, uniform execution template. Build pods where top-down strategic planners work side-by-side with bottom-up anomaly detectors.

• Build Safe Sandbox Environments: Give your teams dark-launch capabilities—like internal Model Evaluation Gateways—where human stakeholders can benchmark, stress-test, and verify model behavior in real time before full production rollout.

The Takeaway

Automation executes code, but human intuition safeguards purpose.

As we deploy smarter agents and faster CI/CD pipelines, our job as quality leaders isn't to compete with the machine's speed. Our job is to bring sharp, cynical, and cognitively diverse human scrutiny to the arena—ensuring that behind every automated pipeline, there is a human who knows exactly what questions to ask.

The Midnight Migration: What ColdFusion, Red Hat, and 2 AM Bullpens Taught Me About the Social Contract of Code

Long before we were managing FedRAMP-cleared cloud platforms, orchestrating multi-region pods, or leveraging generative AI agents to draft test suites, there was simply the bullpen.

In the early 2000s, tech environments weren't sleek, silent, or remote. They were crowded, loud, bathed in fluorescent lighting, and running on a volatile cocktail of stale coffee and pure adrenaline.

Back then, I was embedded side-by-side with developers in high-stress, fast-paced environments. We were testing code blocks written in ColdFusion, JSP, and raw SQL on custom-built Red Hat Linux boxes and WAMP stacks. Deployments didn't happen with a slick CI/CD pipeline or a single click on a dashboard. They happened at 2:00 AM. They meant physically moving rack servers, manually commenting out configuration files in text editors, and running cutovers where a single unindexed query or misconfigured .conf file could bring down an entire data center migration.

In those early days, I wore my motto like heavy armor: Quis custodiet ipsos custodes? (Who watches the watchers?).

I was the kid from Santa Monica College who minored in philosophy just to keep the tech space honest. I believed my sole purpose in the bullpen was to be the unyielding wall—the cynical eye that caught the fatal edge cases before production did.

I thought code was just logic executing on hardware. But during those bleary-eyed, midnight deployments, I stumbled onto a deeper truth: Code isn't just data. It’s a social contract.

The Social Contract of Systems

When software breaks in a way that forces an end-user—whether a loan officer in a mortgage portal or a gamer on a live server—into a clunky "workaround," it isn't just a technical glitch. It’s a breach of trust.

A software bug represents a breakdown in the system’s underlying agreement between product, engineering, and the human being on the other side of the screen.

For years, I believed that guarding that social contract meant pointing out every single anomaly, every edge-case deviation, and every minor flaw. I treated every bug like a 5-alarm fire. But as anyone who has ever survived a data center cutover knows, if everything is critical, nothing is.

Which brings us to the most dangerous file in any system architecture: The Configuration File.

Configuration Management as Neurodivergent Masking

In systems architecture, Configuration Management determines how an application behaves within a specific environment. The raw compiled code might be identical, but the file tells the application how to present itself to the world:

• In Local/Sandbox, debug logging is wide open, verbose, and messy. Errors are thrown freely because the environment is meant for raw processing.

• In QA/Stage, constraints are introduced. API endpoints are locked down, mock services are stubbed, and the build is sanitized.

• In Production, security headers are strictly enforced, SMTP ports are guarded, and only optimized, quiet execution is exposed to the public.

Deploying code with the wrong configuration causes immediate, catastrophic friction. If you accidentally point production traffic at a local debug config, you flood the server with noise, crash the pipeline, and expose vulnerabilities.

It took me twenty years—and a late-in-life dual diagnosis of ADHD and Bipolar Depression—to realize that configuration management is the technical equivalent of neurodivergent masking.

The Cost of the Wrong "Environment Config"

For decades, my internal baseline engine ran at hyper-speed. Growing up as an adopted middle child, my early rule for emotional survival was simple: be quiet, or be punished. In high-stress bullpens, that survival rule morphed into hyper-focus and masking.

I was constantly reconfiguring my personal internal settings depending on which "environment" I was stepping into:

• The "Corporate Leadership" Mask: Dialing down the internal ADHD noise, suppressing side-tangents, and forcing top-down communication patterns to match neurotypical expectations.

• The "Bullpen Firefighter" Mask: Dialing up adrenaline and bottom-up detail processing to catch micro-anomalies in complex data paths.

Masking is essentially running a high-overhead runtime script just to fit the local environment. It requires immense cognitive load. And when your personal "configuration file" is misaligned with your actual internal baseline, burnout isn't just possible—it’s guaranteed.

Stop Writing "Bugs That Don't Matter"

When you live with an unmasked or misconfigured neurodivergent mind in a traditional workplace, your hyper-focus can easily trap you in what I call "the Squeaky Wheel Trap." Because your brain processes bottom-up details faster than context, you begin logging everything. You point out every missing pixel, every legacy process flaw, and every minor inconsistency in team dynamics.

In SQA, this translates to logging 200 low-severity bugs right before a major release deadline. In leadership, it translates to picking every battle until your team stops listening to you altogether.

You end up spending all your energy logging bugs that don't matter.

To actually matter in the workplace—and to build a sustainable career without burning out your internal system—you have to shift from reactive bug-pointing to Quality Intelligence:

1. Understand the Environment Constraints: Stop expecting a Sandbox environment to behave like Production. Learn the specific business context, emotional capacity, and operational risk of the team you are working with today.

2. Prioritize the Social Contract, Not the Semicolon: A minor UI alignment bug in a low-traffic internal tool doesn't breach user trust; a silent failure in a core data pipeline does. Save your political capital and internal bandwidth for the failures that actually break trust.

3. Audit Your Own Baseline Config: Stop trying to rewrite your core source code (your neurodivergence). Instead, manage your configuration intentionally. Be radically candid about your capacity limits, set Work-In-Progress (WIP) limits on your time, and drop the defensive armor.

From Watching the Watchers to Protecting Them

In those 2 AM ColdFusion bullpens, I thought being a great Watcher meant having the sharpest eye for broken code.

Today, as a leader engineering AI-augmented Quality Intelligence frameworks, I know better. The ultimate evolution of leadership isn't building a bulletproof automation pipeline or maintaining a 100% test-pass dashboard.

It’s about recognizing that systems—both software and human—are fragile when they are misconfigured. It’s about building a culture where engineers don't have to wear exhaustion like a badge of honor, and where quality isn't an adversarial gatekeeper, but a shared agreement.

If we want to watch the watchers, we have to start by making sure their configuration allows them to thrive.

Over to You:

How much energy is your team spending hunting "bugs that don't matter" versus protecting the core social contract of your product? How do you manage your own internal "environment configs" at work? Let’s discuss in the comments. 👇

Mechanics of undiagnosed neurodivergence in a high-velocity tech environment operating on a raw, volatile currency: immediate dopamine.

In the early game-testing bullpens of Activision and Troika, that dopamine loop was simple, primitive, and highly addictive. It was the thrill of the hunt. You sat shoulder-to-shoulder under flickering CRT monitors, un-medicated and hyper-focused, waiting for the exact sequence of technical friction that would break an engine wide open. Pushing a high-severity ticket into the tracker and watching a multi-million-dollar build collapse felt like an absolute jackpot. It was a reactive, post-development high—a "monkey pointing at a problem" after the code had already left the sandbox.

But as the infrastructure scaled from local game discs to globally distributed cloud ecosystems, that old loop became unsustainable. Reactive quality control wasn't just slow; it was an operational liability that burned out human networks.

The real transformation of my career occurred when I intentionally shifted the pipeline—and my own internal cognitive engine—from reactive bug hunting to a highly systematic Lean-Left paradigm.

Shifting the Dopamine Loop

True Quality Intelligence (QI) meant migrating the source of that intellectual fulfillment. I realized that the satisfaction of catching a broken deployment at midnight couldn't compare to the strategic triumph of architecting a gateway that refused the anomaly entirely during the design phase.

We decoupled from the legacy relics of traditional QA and anchored our engineering throughput to objective, industry-standard DORA Metrics (DevOps Research and Assessment). By measuring organizational velocity and platform stability, the dopamine loop successfully shifted from finding the break to optimizing the flow:

• Deployment Frequency: Transitioning from sluggish, hyper-cautious monthly rollouts to rapid, continuous, automated iterations. The new high wasn't stopping the release; it was clearing the launchpad safely multiple times a day.

• Lead Time for Changes: Leveraging Generative AI engines (like custom Gemini CLI tools) to slash manual test case authorship effort straight down from 508 hours to 251 hours. Bypassing bureaucratic bottlenecks to compress the timeline from code commit to production verification.

• Change Failure Rate: Shifting the "What Could Go Wrong?" muscle all the way to the research phase, driving the production defect rate down to near zero by establishing a shared social contract of code before a single line was written.

• Time to Restore Service (MTTR): Replacing chaotic, adrenaline-fueled troubleshooting sessions with proactive telemetry curves and automated fallback safety nets.

The Velocity Ledger

By leveraging algorithmic efficiency to handle the repetitive baseline coverage, our human capital was reallocated to where it delivered the highest strategic impact: human-in-the-loop cynical scrutiny.

Operational Dimension	The Old Bullpen Loop (Reactive)	The DORA-Driven QI Loop (Predictive)
Dopamine Source	Splitting open a broken build post-dev.	Engineering lean, friction-free deployment velocity.
Methodology	Rigid, post-development gatekeeping.	Lean-Left Specification Driven Development (SDD).
Primary Metric	Cumulative defect counts in Jira/Mantis.	Optimized Delivery Velocity & Systemic Stability.
Cognitive State	Masked hyper-focus fueled by crisis.	Psychological safety protected by automated guardrails.

"A workaround is just a polite word for an engineer cheating a system that refuses to accommodate human momentum. True Quality Intelligence means architecting environments where the right path is naturally the most efficient path."

At forty-nine, the mantra Quis custodiet ipsos custodes? remains hardcoded into my system, but the watchtower has changed. The internal engine no longer runs hot on the chaos of an active firefighting cycle. Instead, it thrives on the elegant, quiet precision of a high-availability architecture where the pipeline is lean, the pods are protected from burnout, and the system is stable enough to let the people step out of the dust of the arena and simply breathe.