Controlled review preview Not approved for public release
Nick Ferguson IT Pro Direct

Nick Ferguson · IT Pro Direct · OpenAI DevDay 2026

I build practical AI systems for messy, evidence-heavy work.

Through IT Pro Direct, I work on AI implementation that preserves evidence, keeps human judgment in control, respects security boundaries, and follows through into workflows people can actually review and use.

View Research Intelligence See Selected Work

Flagship case study

Research Intelligence

Evidence-grounded workflows for AI-assisted research and decision making.

Why it exists

The investigation can become harder to manage than the model.

In fast-moving AI-assisted research, evidence, uncertainty, failed paths, decisions, and re-entry context accumulate quickly. A durable research state helps keep the work inspectable.

The method

Canonical evidence stays authoritative.

Analytical state and model synthesis remain derived and reviewable. Provenance travels with the work, uncertainty remains explicit, and human authority stays at the consequential decision boundary.

Research Intelligence workflow from human authority through structured research state, models and sources, evidence and provenance, to decision support.
A reviewable path from human intent to evidence-backed decision support—not an autonomous decision pipeline.

What the evidence currently supports

Two studies, mixed results, one durable operating lesson.

Study 001

System emerged under pressure

A local analytical system grew alongside live model-behavior research. The reviewed source package is complete, independently reviewed, and human approved.

Study 002

The method was tested more deliberately

Seven preregistered hypotheses closed with five partially supported, one insufficient evidence, one unresolved, and zero fully supported results.

5Partially supported
1Insufficient evidence
1Unresolved
0Fully supported

Formal hypothesis classifications, not an overall score or success rate.

Two-study Research Intelligence summary showing Study 001, Study 002, the four Study 002 classifications, and the primary cross-study lesson.
A visual summary of the reviewed evidence state; the classifications remain mixed by design.

Primary cross-study lesson

The operating method carried forward more clearly than the software engine did.

What is not claimed

The current evidence does not establish measured efficiency improvement, causal performance gains, generalized transfer, team handoff, enterprise readiness, or ROI. Those remain open questions rather than implied wins.

Supporting proof · Agentic systems

Three in the Loop

A WebMCP research workbench exploring how a human, an agent, and the live web divide authority without losing provenance or review.

  • Human authority sets goals, constraints, and consequential decisions.
  • The agent orchestrates work while sources and provenance stay inspectable.
  • Review remains a system boundary, not a final-minute formality.
Open the live workbench
Three in the Loop authority diagram connecting human authority, agent orchestration, and live web sources around provenance and review.
Authority is divided; provenance and human review connect the system.

Selected work · Practical implementation

Different proof for different conversations.

The implementation identity stays consistent. The evidence changes with the conversation—adoption, field operations, commercialization, research, security, or agentic systems—and remains bounded by what can be verified and shared.

Conceptual practical AI implementation graphic spanning onboarding and adoption, field operations, and commercialization.
Concept illustration of implementation contexts—not evidence of measured outcomes, customer ROI, deployment scale, or partner status.

Ferguson · Approval-gated evidence

Real-world implementation context

Selected Ferguson-related work can provide a separate evidence track for onboarding, practitioner workflows, and commercialization. Named-company claims and artifacts stay withheld until approval, verification, and sanitization are complete.

Build Week · Completed product

My Crafted Career

A career decision-support product built during Build Week around a real claims-career problem—useful proof of focused product execution, not a scale claim.

Security + systems · Technical depth

Evidence-aware technical work

Cybersecurity, cloud, network, evaluation, and operating-system projects add the technical and security context required for responsible implementation work.

DevDay objective

Find where this capability is genuinely useful.

I’m interested in conversations with implementation teams, partners, builders, and operators working through the gap between an impressive AI demo and a workflow people can trust, use, review, and improve.

Connect on GitHub