Innovation creates possibilities. Systems create businesses.
Operating Autonomous Fleets
A Product Operator’s Runbook explores how organizations assign authority, budget work, accept results, and recover when autonomous agents stop. It connects agent execution to ownership and the cost of a completed business outcome.
The runbook includes a four-part essay series in preparation, editable reference templates, and a published field guide on building coding-agent evaluations you can trust. The proposed controls require review and implementation in the system that runs the work.
Explore the runbook · Use the templates · Read the eval guide
Platforms, intelligence, product, and business
My work connects product strategy with the architecture and operating decisions required to deliver it. I work across four related areas.
| Pillar | Questions I work on | Systems and disciplines |
|---|---|---|
| Platforms | How can teams build on a capability and depend on it? | Developer ecosystems · APIs · AI platforms · Cloud architecture |
| Intelligence | How does data support decisions and accountable action? | Agent systems · Analytics · Data foundations · Decision engines |
| Product | What customer problem is worth solving, and can the economics work? | Product architecture · Platform economics · API monetization |
| Business | Who owns the outcome, and how does the organization learn? | Operating models · Autonomous organizations · Business experiments |
Systems
DRAWING SET P-100
These projects cover agent operations, analytics, enterprise integration, and data workflows.
| REF | COMPONENT | FOCUS | SOURCE |
|---|---|---|---|
01 · OPS |
watu-system | Agent roles, shared skills, and integrations for coordinated agent work. | Private |
02 · REF |
Operating Autonomous Fleets | Operating-model reference: authority, budgets, acceptance evidence, and recovery, with editable reference templates. | Public reference |
03 · EVAL |
Coding-Agent Evals You Can Trust | Decision-grade evaluation protocol for repository agents: two-sided grader gates, repeat variance, harness controls, and blinded maintainer review. | Public guide |
04 · AUTH |
agent-credential-broker | Credential access and authorization for agent systems. Under development. | Private |
05 · ANALYTICS |
analytics-agent-v2 | Analytical agents with interactive queries and visual results. | Private |
06 · MODEL |
analytics-llm-finetuning | End-to-end LoRA fine-tuning, schema validation, and evaluation pipeline for tool-calling analytical models. | Private |
07 · API |
tuma | API gateway and service-account setup for machine-to-machine integrations. | Private |
08 · APP |
streamExcel-light | Excel conversion application with a preview and editing step before JSON or XML export. | Public |
09 · DATA |
mssqsl-postgres-dlt-pipeline | Incremental data pipeline from SQL Server to PostgreSQL as a foundation for downstream analytics. | Public |
Theses
These principles guide how I evaluate products and the systems that support them.
- Architecture is a business decision. Evaluate it against customer needs, delivery cost, and the ability to change.
- A product strategy starts with a customer problem and a reason to pay for the outcome.
- APIs become products when people can depend on their contracts, support, and economics.
- Measure the full cost of an accepted business outcome, including review and recovery.
- Give autonomous work an accountable owner, a spending limit, and evidence for acceptance.
- An analytics agent depends on the SQL models and schema contracts feeding its context.
- Infrastructure should make products easier to deliver and operate.
- Use bounded, testable state transitions where decisions require predictable control.
About
I’m Patrick Roch, VP, Platform Product Management & Solution Architecture. My work spans developer platforms, analytical systems, cloud infrastructure, and governed agent runtimes. I connect those technical choices to product delivery, platform economics, and operating responsibility.
Washington DC Metro / Global
Build → Operate → Learn → Explain