focus: making messy operational systems legible enough to turn into software
Sam's work clusters around business systems where several plausible sources disagree and somebody still has to decide what the software should do. He moves between client intent, data semantics, product behavior, and live runtime state. The recurring expertise is not a particular framework. It is turning operational ambiguity into a system that can be inspected, used, and corrected.
Best fit: product-technical lead for operational software, SaaS replacement, or AI-enabled internal systems. Most useful when the hard part is mapping a real business process, deciding which source is authoritative, and staying involved through implementation and live use.
Less like: a conventional individual-contributor programmer hired to receive a settled specification and spend most of the project hand-authoring one subsystem.
The primary development environment. Codex supplies the longer historical record; Claude Code appears heavily in recent product work, reviews, and parallel execution.
Frequently moves outside the code editor to inspect authenticated applications, cloud consoles, email, CRM state, and live API responses.
Uses parallel work for separable tasks and independent model roles for review. The pattern is most developed in audits, implementation batches, and high-risk changes.
Pi, Gemini CLI, Cursor history, MCP servers, local-model experiments, scheduled agents, and reusable skills appear as supporting tools rather than the main environment.
Routes higher-judgment design and synthesis work differently from mechanical implementation, and has reviewed whether that routing actually saves money.
The most common application stack in the record: dashboards, internal tools, membership software, reporting interfaces, and public sites. Work is primarily directed and reviewed through agents.
The surrounding language environment for most web work. The history supports architecture, debugging, and review familiarity more clearly than unaided code authorship.
Pages, Functions, D1, KV, deployment bindings, and token-permission diagnosis appear repeatedly in current projects.
Vercel hosting and scheduled work; GCP Cloud Build and Cloud Run with staging-to-production promotion. Comfortable operating these stacks without evidence of specialist cloud architecture depth.
Appears as application infrastructure alongside D1 and hosted databases. Less evidence of deep query optimization or database-engine specialization.
REST APIs, OAuth, transactional email, issue trackers, CRM and HR systems, browser automation, webhooks, and data exports recur more than any one backend framework.
Client delivery, account management, staffing, project evidence, meeting systems, scorecards, issue hygiene, and executive reporting.
CRM, HR, payroll, finance, identity, permissions, historical backfills, source ownership, and safe production synchronization.
Booking rules, billing behavior, e-signature, membership flows, mobile packaging, and operating tools for a private-club environment.
LLM application integration, agent-enabled delivery, model routing, document analysis, browser agents, and operational adoption.
Aviation training systems, education-sector software advisory, and private-equity document or reporting workflows appear repeatedly but with less volume than business operations.
Broader evidence in product definition, business process, integration semantics, and live operations. Weaker evidence of personally hand-authoring complex code or specializing deeply in one backend or infrastructure layer.
Much closer to the implementation: data models, interface behavior, runtime state, access rules, deployment, and verification all remain inside the working scope.
Stronger evidence in applying models inside products and operating workflows. Much less evidence in model training, formal evaluation research, embeddings infrastructure, or novel ML methods.
More product and organizational judgment, especially around source authority and client-owned decisions. Less centered on optimizing one automation platform or integration tool.
Operational internal tools, SaaS replacement, cross-system reconciliation, AI-enabled service delivery, and product work with an unsettled business process.
Security-critical infrastructure, sophisticated data engineering, high-scale backend systems, native mobile, or deep database work.
Model training, research ML, low-level systems, novel algorithms, or a role evaluated primarily on unaided coding speed.
Coverage includes 157 substantive Claude Code sessions from June 8 to July 9, 2026; 183 Codex sessions from August 19, 2025 to July 9, 2026; six Pi sessions; 23 Gemini CLI sessions; and limited Cursor metadata.
Client, company, and personal identifiers were abstracted. The profile uses local session evidence only. Absence of evidence is not evidence of absence.