Browser investigation
When Saved does not mean saved
A deliberately broken example shows what to inspect when a success message hides a failed request.
Follow the Jevitate example →Product & engineering dossier New York / Remote
Product strategy and engineering, working together.
I connect product strategy with engineering so teams can make deliberate choices about what to build, when to build it, and how to keep it dependable.
Unclear direction and requirements can turn engineering effort into rework. I focus on keeping the customer's problem, the business decision, and the engineering work connected—from the first assumption through requirements, handoffs, and verification.
I've led both product and engineering teams at artnet and OutboundLabs, and held Staff Engineer roles at Distributed Spectrum and Qrypt. I stay hands-on with architecture and software development, balancing product needs with reliability, security, maintainability, and operating cost.
Today, I apply that approach through MC Product Labs and open-source tools that help clarify work, coordinate execution, and check the result.
Examples and evidence
Work you can inspect.
Follow a concrete example, inspect the tools, or read an attributed account of the work.
Delegated coding
Keep the handoff reviewable
Inspect how Junior Agent scopes work and keeps independent checks and results available for review.
Explore Junior Agent ↗Platform modernization
Read the external account
Persistent's case study describes artnet's migration to Google Cloud with MediaAgility.
Read the case study ↗Product company
Products built and stewarded through MC Product Labs.
Through MC Product Labs, I start with a narrow customer group and a painful problem, then test the assumptions that could change the next investment. The aim is to avoid building before the need is understood and to carry what we learn into clear engineering work. The company blog covers the product work; my personal writing covers engineering and open-source projects.
Preveti
Vet before bet—challenge product assumptions, examine alternatives, and authorize evidence before committing engineering time or capital.
- Assumption testing
- Simulation + human studies
- Decision records
Allumata
Engineered content for agencies—structural specs, validation rubrics, and signal loops so copy is manufactured to an outcome, not guessed at.
- Structural analysis
- Side-by-side audit
- Section-level attribution
Attestrue
Independent evidence when proof matters—signed records, policy-version evidence, and corroboration for consent and compliance events.
- Signed records
- Policy-version evidence
- Proof anchors
Licenz
Enterprise airgap licensing in a 2MB Rust binary—offline-first validation, hardware binding, and an auditable open-source client.
- Offline-first
- Air-gapped support
- MIT-licensed client
PubOps
The operations plane for public posture—inspect legacy claims, authorize repositioning moves, and govern required expressions across channels.
- Posture-gap review
- Human-authorized moves
- Evidence-bounded
FrontRails
A suite of MCP tools for teams who want LLM speed without drift, rework, or broken flows—intent, structure, security, and UX aligned to what actually ships.
- MCP tool suite
- Local CLI control layer
- System-level alignment
Open Source · Junior Agent
Scoped coding tasks.
Reviewable results.
Junior Agent delegates bounded coding work to lower-cost models using acceptance criteria, isolated Git worktrees, independent checks, and bounded repair attempts. I actively use it across my own projects.
Explicit task contracts
Scope and acceptance criteria define what each delegated task must deliver.
Independent checks
Verification and bounded repair attempts produce evidence for review.
Reviewable handoffs
Changes, check results, model usage, and unresolved work remain visible to the reviewer.
Praxec tooling
Earlier MCP/CLI tools for critical-path planning, House of Quality analysis, and failure-mode assessment support task planning and review.
Earlier open source · Praxec
Safe AI requires deterministic boundaries.
Praxec is an open-source agentic harness—hosting workflows, enforcing policy gates, and retaining control over state, tools, approvals, and evidence while models do probabilistic work.
Explore the Praxec organization ↗The harness
Hosts agentic workflows with legal transitions, durable policy gates, human approval, and audit trails—authority in code, not model interpretation.
Cognitive architectures
Structured, governed approaches for discovery, analysis, FMECA, implementation, and verification.
Deterministic MCP tools
Open-source utilities—CPM planning, elicitation, scientific process, cross-matrix analysis, corpus work, and operational tools—that plug into the harness.
FrontRails integration
Praxec’s MCP and CLI utilities also support Junior Agent task planning and review, including critical-path planning, House of Quality analysis, and failure-mode assessment.
Family Study: Temporal Runtimes
Event-processing for distributed systems.
The tflo family explores embeddable temporal event processing across Rust, WASM, React, and browser environments—with shared models and deterministic behavior as design goals.
Explore tflo repositories ↗Runtime
tflo-core
The embeddable Rust engine and shared temporal foundation.
Interface
tflo-react
Declarative temporal state for React applications backed by the shared runtime.
Interaction
tflo-browser-events
Browser-event capture and signals designed around durable identifiers and pluggable sinks.
Open Source · Jevitate
Autonomous browser testing with a deterministic spine.
Jevitate records real browser sessions into typed Journeys over Playwright, then replays, explores, load-tests, and reviews them from a single CLI or MCP server. Autonomous runs are driven by TypeSafe’s Jev and bounded by hard guardrails—model decisions are hypotheses; runtime evidence decides pass or fail. MIT-licensed, TypeScript, Node 20+.
Record → replay
Capture a real flow by demonstration into a deterministic, typed Journey—then parameterize, promote, and load-test it against an authorized origin.
Goal-directed exploration
Drive to a natural-language goal with autonomous exploration; success is judged by an independent assertion, never the model’s say-so.
Adversarial & UX review
Bounded misuse against a trusted hard-signal defect oracle, plus ranked, cited, evidence-anchored usability findings.
MCP server, safety-first
Exposes only an allowlisted tool surface for agents; every run is origin-bound and budget-capped, and credentials are never sent to a model.
Open Source · opv
Secrets delivery with explicit controls.
opv (1password-vault) is an open-source Rust CLI connecting 1Password-managed secrets to application runtimes and CI. 1Password remains the source of secret values; opv coordinates validation and delivery through official CLIs.
Typed validation
Declare key names, IDs, and validation rules. Reporting identifies problems without displaying secret values.
Plan before sync
Review proposed changes and bind synchronization to a plan. Deployment and pruning require explicit controls.
Runtime and CI targets
Supports Fly.io, Azure Key Vault and Container Apps, Kubernetes, External Secrets Operator, and GitHub Actions.
Automation and assistants
Structured JSON reports, stable error codes, and an agent skill support scripted and AI-assisted workflows. Distributed as prebuilt binaries and through npm; MIT licensed.
Engineering / Attributed Evidence
Engineering &
Attributed Evidence
Breadth is useful when it becomes pattern recognition: across public companies, billion-dollar enterprise software, regulated domains, and early-stage teams.
Public-company CTO leadership
Hands-on platform modernization
Owned modernization of a decades-old monolithic platform, actively designing, planning, managing, and helping execute the transition toward cloud architecture. Led collaboration with MediaAgility on the phased migration to Google Cloud.
Read external case study ↗Billion-dollar enterprise · Medidata
Flagship-product architecture and portfolio repair
Architected and repaired flagship products and led technical turnarounds for some of the weakest products in the portfolio—work requiring product judgment, delivery discipline, and hands-on technical depth.
Program and delivery governance
Critical-path discipline for consequential work
Applies formal project planning, including CPM, to expose dependencies, sequence work, and control delivery risk—and builds tools that bring the same discipline to agentic coding.
Working method / 03
Conceive the product.
Steward the system.
Discovery, boundary analysis, critical-path planning, implementation, and verification form one accountable delivery loop. The artifacts change; the discipline does not.
Discover→Model risk→Plan the path→Ship→Verify
Decisions and field notes
Explore the work behind the approach.
Read practical examples of product decisions, engineering tradeoffs, and software verification. My professional background and project history are available in the résumé.