A vast old tree shelters a lit workshop under a blue night sky — the Broken Branch lab.

Capability dossier · public field copy 01

Ben Schippers Put me on the hard problem.

I turn ambiguous AI ideas into shipped, testable systems. Product judgment and program discipline stay human; an agent fleet supplies execution; evidence decides what survives.

Microsoft cloud engineer · product and program operator · Atlanta, Georgia

Bring me a hard problem Inspect the evidence

Public evidence only. Client-sensitive work, credentials, and fleet internals stay private on purpose.

open the file

01 / Assignment fit

Where I create leverage.

Three kinds of work. One operating pattern: find the load-bearing problem, make the system legible, and ship it all the way through.

AI product & platform

Turn a model capability into a coherent product, platform surface, or developer workflow—with a real user, an explicit quality bar, and an honest path to production.

Best fit: agents, developer tools, applied AI, enterprise platforms

Technical program leadership

Run broad, parallel technical portfolios without losing the critical path. Build the mechanisms that make risk, ownership, handoffs, and executive decisions crisp.

Best fit: infrastructure, cloud, security, multi-team delivery

Forward-deployed & special projects

Enter an evolving problem space, learn directly from users and operators, then own the complete solution—from first framing through the production handoff.

Best fit: high-consequence deployments, zero-to-one systems, field work

02 / Selected proof

Three systems.
Three different kinds of evidence.

The portfolio is larger. These are the three systems that best expose the operating range: organizational leverage, commercial product judgment, and rigor under uncertainty.

Case 01 · organizational leverage

Microsoft · from silos to signal

A shared intelligence system for enterprise AI adoption problems no single product team could see.

Problem
A major enterprise AI rollout was accelerating across product lines with no shared view of the adoption walls customers were hitting.
My decisions
Assemble the cross-functional forum, deploy real-time case analytics, and turn fragmented support signal into a durable engineering-priority loop.
Direct result
The system surfaced 76 blockers and directly unblocked 7,380 users.
Program outcome
The broader work helped the organization add approximately 94,000 seats. That is a shared program result; my lever was the pipeline and the forum that ran on it.

Case 02 · commercial product

VibeCrafting

AI physical-product design aimed at fabrication artifacts—not an attractive picture and a shrug.

Problem
Generative tools can make physical ideas look plausible without producing the structured information required to fabricate them.
My decisions
Define the product around cut lists, shopping lists, exploded views, stock-aware ideation, and printable connectors that replace difficult joinery.
Agent execution
Implement the full-stack workflow, parametric geometry, OpenSCAD exports, and iterative product surfaces under that specification.
Current evidence
The live product and public sample gallery expose the output today. Customer and revenue metrics stay blank until there is a number worth publishing.

Case 03 · rigor under uncertainty

Erdős

An AI-math system where the method—not the model’s confidence—is the product.

Problem
An AI doing mathematics can sound convincing while being quietly wrong, and most systems discard the reasoning trail.
My decisions
Separate a deterministic Calculator from a provenance Ledger. Record every branch, dead end, tool call, confidence layer, and verification claim.
Quality gate
“Verified” is earned only by a real Lean subprocess exit code. Replay must reconstruct the run from cache without re-executing it.
Evidence
The first replay caught a real ULID ordering bug. The first Lean-backed result reports no axioms; a separate headline hypothesis remained explicitly unproven.

03 / Operating model

Directed, not hand-typed.

I use agents as an execution layer, not an accountability shield. Judgment, system design, acceptance criteria, and the release decision stay human.

  1. 01Frame

    Find the real user, constraint, and failure condition.

    Human lead
  2. 02Decompose

    Turn ambiguity into bounded work with explicit interfaces.

    Human + agents
  3. 03Execute

    Run parallel implementation, research, and inspection loops.

    Agent fleet
  4. 04Verify

    Use tests, deterministic tools, browsers, and receipts—not summaries.

    Evidence lead
  5. 05Release

    Resolve the seams, own the tradeoffs, ship the whole system.

    Human accountable

The line I will not blur

Agents may produce the work. I remain accountable for the specification, the evidence, and what gets called finished.
Read the operating philosophy

04 / Record

Scale before the fleet.
Velocity after it.

The current lab shows range and execution speed. The earlier career shows what happens when the same operating instincts meet enterprise scale.

Enterprise operating record

5 yearsas a Microsoft Senior Program Manager across Copilot, Graph, Windows 365, and Teams Devices

8product lines in an internal platform portfolio spanning signal, routing, self-service, and quality measurement

95+features shipped through a rebuilt signal-to-engineering pipeline

433Kusers in a crisis transition led with near-zero churn

3 → 150agents in the Premier Engineering program built from pilot to scaled operation

60K/yrincidents handled by that scaled program

Career-record claims shown with shared program outcomes qualified separately. Supporting detail is available in conversation.

Inspect the full career cross-section

Public fleet pulse

Git-derived, timestamped, and linked to public receipts. If the feed gets stale, it disappears instead of bluffing.

Inspect the public work

Open-source receipt Career Compass turns a ten-month job-search workflow into a TypeScript MCP server with durable career context and a local pipeline dashboard. Read the source

05 / Sensitive work

Clear in public.
Disciplined in private.

The lab publishes methods, artifacts, null results, and receipts. It does not publish credentials, client-sensitive information, or operational internals that do not belong on the open web.

For public-sector, security, or other high-consequence work, I would rather discuss the actual constraints directly than decorate this page with claims I cannot responsibly prove here.

End of public file

Bring me the hard problem.

If you are building ambitious AI infrastructure, a product that must survive contact with reality, or a special project without a clean playbook, I would like to hear about it.

Email Ben LinkedIn Download the brief