The Playbook Library.

JUN 26 · 20264 min readExecution · Deployment

FM-02 · FIELD MANUAL

From Pilot to Production

The graveyard of enterprise AI is full of successful pilots. The gap between a demo and a deployed system is organizational, not technical — here is how to cross it.

JUL 1 · 20264 min readStrategy · ROI

FM-03 · FIELD MANUAL

The High-ROI Use Case Filter

Every team has forty AI ideas. Four are worth building. A five-question filter for finding them — before you spend a dollar on the wrong ones.

MAY 20 · 20264 min readGovernance · Security

FM-04 · FIELD MANUAL

AI Governance for Mid-Market Companies

You don't need an AI council, a 40-page policy, or a chief AI ethics officer. You need one page of rules, three risk tiers, and an audit trail.

JUL 3 · 20266 min readStrategy · Execution · ROI

FM-05 · FIELD MANUAL

The Workflow Audit Manual

Your best automation candidates are hiding in plain sight — in inboxes, spreadsheets, and one veteran's Tuesday routine. A one-week audit that surfaces and ranks them without a consultant in sight.

JUL 6 · 20265 min readExecution · Strategy

FM-06 · FIELD MANUAL

The Build Pod Operating Manual

A monthly build lane is only as good as the operating rhythm around it. How to run your side of a pod engagement — intake, approvals, demos, acceptance — so every credit compounds.

JUL 8 · 20266 min readAdoption · Design · Execution

FM-07 · FIELD MANUAL

The Adoption Playbook

Working software that nobody uses is the most expensive kind. The rollout protocol that turns a shipped automation into a habit — skeptic recruitment, draft mode, miss logs, and the graduation vote.

JUL 5 · 20266 min readVendors · Strategy · Security

FM-08 · FIELD MANUAL

The AI Vendor Diligence Manual

Every tool is AI-powered and every demo is flawless. A structured diligence protocol — the paper trail, the proof-of-work test, and the reference call script — for buying AI without buying a story.

JUL 7 · 20265 min readExecution · Observability · Governance

FM-09 · FIELD MANUAL

The Automation Operations Manual

Launch is the midpoint, not the finish line. The operating rhythm that keeps automations healthy for years — the weekly review, the drift watch, the incident play, and the quarterly health check.

JUL 9 · 20265 min readData · Strategy · Engineering

FM-10 · FIELD MANUAL

The Data Foundations Manual

Not the data lake pitch — the minimum viable data readiness for one workflow at a time. What "reachable" actually means, the access patterns that unblock builds, and the foundations that pay off in weeks.

JUN 28 · 20266 min readEngineering · Retrieval · Data

Schematics

technical whitepapers · code included

The engineering underneath the workflows: retrieval architecture, approval gates, extraction pipelines, observability, and security — in the weeds on purpose, with copyable code.

SCH-01 · SCHEMATIC

Retrieval Architecture for Internal Knowledge Assistants

Why naive vector search plateaus at "demo good," and the production architecture that beats it: structure-aware chunking, hybrid search with reciprocal rank fusion, reranking, and citations that survive an audit.

JUL 2 · 20265 min readEngineering · Governance · Architecture

SCH-02 · SCHEMATIC

Approval Gates: Engineering Human-in-the-Loop Agent Workflows

"A human approves it" is a state machine, an idempotency contract, and an audit log — or it's a liability. The engineering behind approval gates that hold up under retries, race conditions, and auditors.

JUL 6 · 20265 min readEngineering · Data · Design

SCH-03 · SCHEMATIC

Structured Extraction: Documents In, Validated JSON Out

Extraction is the workhorse of document automation — and the difference between a demo and a system is validation loops, calibrated confidence, and routing. The full pipeline, with code.

JUL 4 · 20266 min readEngineering · Observability · Architecture

SCH-04 · SCHEMATIC

Tracing and Evals for LLM Workflows in Production

You can't improve what you can't replay. The observability layer every production LLM workflow needs — traces, cost accounting, offline evals, and drift detection — with the code to start.

JUL 8 · 20266 min readSecurity · Engineering · Architecture

SCH-05 · SCHEMATIC

Defending Agent Workflows Against Prompt Injection

Any agent that reads untrusted content and can take actions is running someone else's instructions — unless you engineer otherwise. The defense-in-depth architecture, from tool policy to canary tokens.

DSP-01Why Most AI Projects StallIt's not the model. It's not the data. The stall pattern is organizational — and it repeats so reliably you can set your roadmap by it.

Dispatches

long-form essays · strong opinions

APR 30 · 20264 min readExecution · Strategy

DSP-02Modernize Before You Add AIAI layered on a broken workflow makes the mess faster. The highest-ROI "AI project" at many companies is the unglamorous modernization that comes first.

MAY 8 · 20264 min readStrategy · Data

DSP-03Automation Without Losing ControlThe fear is legitimate: automated systems acting invisibly on your behalf. The answer isn't less automation — it's control designed in as a feature, not bolted on as an apology.

JUN 5 · 20264 min readGovernance · Design

DSP-04Secure AI Adoption in Regulated IndustriesHealthcare, finance, and other regulated operators can adopt AI without betting the license — if the deployment respects three boundaries most vendors gloss over.

JUN 19 · 20264 min readSecurity · Compliance

DSP-05Build vs. Buy for AI Workflows: A Decision That Isn't BinaryThe build-vs-buy debate assumes two options. Real AI workflow decisions have four — and the expensive mistakes come from picking the right answer to the wrong layer.

MAY 15 · 20264 min readStrategy · Vendors

DSP-06The True Cost of a Manual WorkflowMost automation business cases undercount by half because they only price the labor. Here's the full ledger — and a back-of-envelope method you can run in an afternoon.

MAY 22 · 20264 min readROI · Strategy

DSP-07Why a Chatbot Shouldn't Be Your First AI ProjectThe default first AI project is a chatbot. It's also the hardest kind to get right and the easiest to abandon. Better first projects share three traits a chatbot lacks.

JUN 26 · 20264 min readStrategy · Design

DSP-08Human-in-the-Loop Patterns That Actually Get UsedMost review queues die of neglect within a month. The difference between an approval gate people use and one they rubber-stamp is interaction design, not policy.

JUL 1 · 20264 min readDesign · Governance

DSP-09How to Write an Automation Request Engineers Can Actually ScopeThe gap between "we should automate this" and a shippable scope is a one-page document. Here's the template — and the two questions that do most of the work.

JUL 3 · 20264 min readExecution · Design

DSP-10Twelve Questions That Cut Through AI Vendor NoiseEvery tool is "AI-powered" now. These twelve questions separate systems that will survive your production reality from demos wearing a pricing page.

JUN 10 · 20263 min readVendors · Strategy

DSP-11You Don't Need a Data Lake to Start with AI"Our data isn't ready" has killed more good automation projects than any technical failure. The readiness bar is real — but it's per-workflow, and it's lower than the platform pitch says.

MAY 29 · 20264 min readData · Strategy

DSP-12Measuring Automation ROI Without Lying to YourselfAutomation ROI numbers are usually either theater or absent. A measurement design you set up before launch — three numbers, honestly counted — beats both.

JUN 14 · 20264 min readROI · Execution

DSP-13Getting Your Team to Trust an AutomationThe technology works and nobody uses it — the most expensive failure in automation. Trust isn't a rollout memo problem; it's a sequence you can design.

JUN 22 · 20264 min readAdoption · Design

DSP-14The Maintenance Bill Nobody Budgets ForAutomations don't fail loudly — they rot quietly. What actually breaks post-launch, what upkeep truly costs, and why the handoff model determines whether value compounds or decays.

JUL 5 · 20264 min readExecution · Strategy

DSP-15Right-Sizing the Model: When Small Beats State-of-the-ArtDefaulting every workflow to the biggest model is like commuting in a freight truck. A pragmatic sizing framework — and why the interesting costs aren't the per-token ones.

JUL 7 · 20264 min readDesign · ROI