I lead operations at Tribunus Health, a healthcare contract-negotiation firm, and I build the AI systems my teams run on: agent systems with real guardrails, automations that replaced hours of weekly manual work, and the governance that decides which of them earn production. In 2026 I completed all three levels of Leland's AI Builder Program and led our AI Task Force's pilot portfolio. Co-author, The New Science of Hiring.
LinkedIn·patricktweeks@gmail.com
Co-author of The New Science of Hiring with Joy Giles, a research-backed guide to evidence-based selection.
Building tools is half the job. The other half is leading an organization into using them well: policy, people, and proof. At Tribunus I founded and led the AI Task Force and wrote the documents the company runs its AI work by.
I assembled and lead our cross-functional AI Task Force, launched at a company kickoff in May 2026. Its mission, in the words I opened with: eliminate the draining, robotic tasks, not replace humans. Free people for the creative, strategic craft they actually enjoy. The Task Force ran a full pilot portfolio through 2026 and graduated its pilots into the functional departments that now own them.
I wrote the one-page North Star & Charter that serves as Tribunus's AI policy: 4 guardrails every AI initiative is judged against: Limit Risk / Protect the House, Gain Efficiency, Unlock Innovation, and Maintain Responsibility. It is the document new pilots are measured against before anything is built.
Every pilot is scored on my 6-dimension, 30-point evaluation framework (Green / Yellow / Red / Stop zones), tracked in a shared pilot tracker, with a hard rule that a score of 1 on any single dimension drops a pilot a full tier, whatever its total. Weak pilots get killed on evidence, not opinions.
I ran the company AI readiness survey, led the kickoff, and took the hard questions head-on (time, billable hours, job scope) with written briefs, including a factual synthesis answering staff environmental concerns about AI. The result: pilots rolled out with training, and the Executive Summary process adopted as a company-wide mandatory standard in September 2026.
Three builds, chosen to show range: return on investment, technical complexity, and organizational leadership.
Took a client deliverable that consumed 6–8+ hours over a week and made it a 20–60 minute process, a ~95% time reduction. Approved as the company-wide standard at Tribunus Health in September 2026.
An agent system in Claude Code that runs my weekly accounts-receivable follow-up at Tribunus: it researches every client with a read-only subagent, drafts the collections notes, and grades its own work, while being physically unable to edit the live sheet or send an email.
The charter and scoring system that governed every AI pilot at Tribunus: a North Star charter with 4 guardrails, and a 6-dimension, 30-point evaluation scorecard with Green / Yellow / Red / Stop zones.
Everything below is in production, ran on a schedule, or shipped as a reusable system, organized by the kind of work, not by when I learned it.
Every Tribunus client account has one current profile, assembled automatically from 4 systems: Drive, Docs, Gmail, and the CommonThread platform. Executive summaries, contacts, and billing history refresh on their own, so any account can be understood in minutes.
A Gemini-based assistant the Tribunus pricing desk uses to diagnose inbound prospects, script discovery calls, and produce standardized pricing quotes with margin guardrails, so pricing no longer depends on one founder's head.
A research system that produced a 248-page competitive intelligence book and a 20-page executive summary: 27 competitors mapped, 473 deal win/loss records analyzed, used directly in Q4 and 2027 strategic planning.
Scoped and sponsored an internal software build with an external development agency that automated Tribunus invoicing and billing workflows and removed a class of manual accounting errors.
My inbox runs on rails: an email-manager skill triages everything, drafts replies in my voice from a guide built out of my own sent mail, and a scheduled run delivers a clean urgent / review / ignore summary every morning before I open my laptop.
A cloud workflow in n8n that triages Gmail and drafts replies, paired with an evaluation sheet where a second model judges the drafts, because an automation nobody measures is just a faster way to make mistakes.
A weekday brief waiting for me each morning, generated by a scheduled task that is only a pointer to one spec file, written to one fixed file at one fixed URL, with history in a text log.
A scheduled routine that compares my plan against the system of record and independent evidence, and reports only the gaps. A plan checked only against itself always passes; this one can't.
A knowledge wiki maintained by AI under a written contract: index first, one entity per file, a source line on every article, lint often, flag before deleting, with /ingest and /lint skills and a scheduled lint.
A skill that audits my entire AI system: every skill, file, and scheduled task: flags gaps, missing instructions, and security issues, and returns a prioritized proposals table. It changes nothing until I approve it.
I run my tennis-elbow and wrist rehab through a personal AI agent with persistent memory: daily check-ins, exercise progression gated on the previous day's pain response, and my whole life (sleep, desk setup, gym load) managed as training load. Week 2 closed with a fully pain-free reassessment.
I set up a fully local alternative to Wispr Flow rather than send my voice to a cloud service: OpenWhispr, an open-source dictation app, running Whisper speech models on my own Windows desktop. Hold a hotkey, speak, and cleaned-up text lands in any app. Audio never leaves the machine.
2 interlocking scheduled agents (a stress coach and a fitness coach) that email me check-ins every evening, parse my replies into persistent case files, track body-composition data against a trajectory chart, and auto-archive their own threads to keep my inbox at zero.
Reusable skills that make every build come out in the right voice and the right look: a writing system that strips AI tells, and design-system skills extracted from live brand files that Claude applies automatically to anything it builds.
AI-driven analysis turned into decision-ready artifacts: a full college-dataset analysis with visualizations and a one-pager, and a data-driven hiring-science deck built from primary research on the Proactive Personality trait.
I teach an agent one real process before it owns anything, cap autonomy in permissions rather than promises, grade outputs with a separate evaluator, and schedule the human review, because a loop nobody reviews is just a faster way to drift. Governance first: every pilot scored, weak pilots killed.
Healthcare operations and payer contract negotiation at Tribunus Health. Before AI, the same instinct: find the recurring manual work, systematize it, measure it. Co-author of The New Science of Hiring, on evidence-based selection.