Richard Pickett

Reading AI's mind.

If you want to read the earlier problems I've solved and why I picked this one next, see Natural Language Compiler and Zygotic AI.

Reading AI's mind: when an agent makes a decision, can we prove the reason it gives is the reason it actually decided?

This is especially important in heavily regulated and high-risk environments, like finance and healthcare.

Describe your app and its constraints in plain English, from architecture to HIPAA, and get an application that’s fully traceable and auditable, built for healthcare, finance, government and defense.

(And just as easy to use if you don’t have all that regulation.)

I built a free toolkit that lets anyone build a fully auditable AI system that enforces compliance - ready for healthcare, finance and government.

Download it and get started at NaturalLanguageCompiler.ai.

• Plain-English intent, including constraints.Describe the app in plain English. Add constraints of your own like hardware, architecture, and company policies as well as any compliance framework, like HIPAA, GDPR, PCI-DSS, and FedRAMP.
• Traceable, auditable application and infrastructure.NLC gives you an application and infrastructure that’s fully traceable and auditable.
• Usable without the regulation load.Just as easy to use if you don’t have all that regulation.
• Intent compiled one gated increment at a time.The AI compiler turns ratified intent into gated, boundary-shaped code from natural-language functionality, requirements, policies, and standards.
• Requirement traced to enforcement when shipped.Trace every requirement to where it is enforced - when shipped.
• Ship only what was intended.Multiple internal audit points confirm what’s intended is what’s built, with final audits to certify the entire build.
• Recompile when the rules change.Update policies, standards, or major technology in intent and recompile when those requirements change.
• Silent drift is eliminated.Problems surface at compile time with evidence - or the build stops.
• Tightly scoped emit; refuse on failed gates.The compiler emits tightly scoped code per step and refuses when gates fail, so defects surface at compile time.
• Audits are a non-event.Every policy, requirement, and standard links to the code that enforces it, ready to export for your next third-party audit.

I don’t just advise on process. I run it.

Zygotic AI is an autonomous AI cluster that executes thousands of operations a day. It learns how to take on new work, surfaces exceptions through andon gates, and keeps quality measurable instead of assumed.

These are my core principles applied at full system scale: process before prompts, fail fast, systems not symptoms, observable by default, improve on purpose.

I solve the hardest problems in applying AI to real business. I build, and teach people to build, always-on agents that keep systems running. They find and fix problems, manage clusters and resolve user issues as they come in, not just write code when asked.

It's even managed by a team of agents. By the way, everything you see about them, from their names, to their photos, and even their roles, were their own choices.

My agents are:

• Staff, not tools.
• Always-on, not one-shot prompts.
• Running the whole system, not single tasks.

For the live view of the cluster, see Zygotic.ai.

Click to see the full architecture on Zygotic.ai

I am an Agentic Systems Process Engineer and Lean Six Sigma Black Belt, building process-first AI systems for production reliability.

I built and run Zygotic AI, an autonomous AI cluster that executes thousands of operations a day and learns how to take on new work.

I lead AI automation and acceleration on the Fixed Income team (FI Quants) at AllianceBernstein.

I’ve also managed systems representing $1B in ARR and previously managed application security across 40 teams.

• Taught myself to program on a Commodore 64.
• 11 years at Idemia in technical roles, including 4 years as Technical Program Manager on a 10-year, $10B federal security contract.
• Lean Six Sigma Black Belt.

We start with a working review of your system: where it breaks, what “good” looks like, and how we’ll measure whether it’s holding up. No boring slides, no hard sales.

I bring the self-improving process, based on built-in quality metrics. You bring the domain knowledge and the decisions only you can make.

Humans stay in the loop for real exceptions, not constant babysitting. The goal is a system you can trust, explain and evolve, not a demo that decays in production.

If this sounds good to you, reach out and let's set up a conversation.

• Making Yourself Irreplaceable In The AI RevolutionAs AI storms through industries like a caffeinated tornado, I’ve stumbled on a gem after years of wrangling machine learning models and coaching startups: most folks are totally lost about what AI can pull off—and what it can’t touch. Think of it as a superhero with a kryptonite catch.Apr 2, 2025 · 11 min read
• My AI Coding Journey: Lessons From The TrenchesMy ChatGPT coding debut was a coin toss — half the time it nailed it, half the time it flung me into a dumpster fire. Sometimes it would recommend packages for my project and give me functional code that worked perfectly. Other times, it would completely hallucinate functions and capabilities those packages didn't have. It was trying to help, creating code to complete my request, but the capability simply wasn't supported. After pointing out these hallucinations, ChatGPT would sheepishly admit Mar 26, 2025 · 6 min read
• My Tech StackBe Opinionated I take all "my OS is better than yours" banter with a grain of salt. (And I give a lot of "Linux is better" banter, lol.) It's true, if all you have is a hammer, everything looks like a nail. We all think "my tool is the best tool for all the jobs." And that's ok. As long as it's informed. Don't be the kid who just learned his first language and thinks that language is the world's best at all things. But do be opinionated about what tools you use and, more importantly, why.Jun 2, 2024 · 15 min read

More posts →