Confidential · The SE Operating Playbook

I don't just sell infrastructure.
I run the system that wins it.

Twenty years in federal and defense taught me that the deal is never decided by the datasheet. It's decided by the conversations, the quote, the relationships — and, increasingly, by who wields AI with the most discipline. This is how I operate.

01 — Portfolio of work

Tools that catch what humans forget.

I build the tooling I wish I'd had. This Data Center Network Designer turns a customer's requirements into a complete, validated design — and surfaces the options engineers routinely miss under deadline pressure: breakout cabling, multi-pod scaling, oversubscription tiers, rack and power layout. Fewer surprises at build time. More trust at quote time.

↳

Why it wins deals: the design that accounts for the breakout cable and the second pod on day one is the design that survives the customer's architecture review — and the competitor's teardown.

02 — Sales as a system

I don't sell products. I run conversations.

The smartest person in the room rarely wins it. The most trusted one does. Every motion below is something I run deliberately, not by instinct.

Don't be the smartest person in the room — earn the room
Ask
→
Listen
→
Reflect it back
→
Earn trust
↺ repeat every meeting
Every audience is different — read it, then choose the motion
Read the room
Demo-led

Builders who trust what they can see running.

Deck-led

Leaders who buy the narrative and the roadmap.

Hands-on POC

Teams who only believe their own keyboard.

Map the room — know your allies and who must be won over
Influence ↑ Disposition →
High influence · skeptical Win them over

The deal lives or dies here. Earn it, or route around it.

High influence · supportive Champions

Arm them. They sell when I'm not in the room.

Low influence · skeptical Monitor

Acknowledge, don't overspend energy. Watch for escalation.

Low influence · supportive Supporters

Intel and reach. They tell me what's really happening.

Leverage every resource — I am never the only person on the deal
The deal
OEMs
Partners
Overlays
Prof. Services
03 — Engineering the quote

The BOM is where trust is won or lost.

Better SKU choices keep a quote competitive and build trust at the same time. I treat the bill of materials as an engineering problem: same outcome, fewer wasted dollars, no features the mission doesn't need.

Same capability, engineered cost
Off-the-shelf BOM Defaults, padded, easy to undercut
→ lower cost · more competitive
Engineered BOM Right SKUs, nothing the mission won't use
Example · memory
24 × 32GB RDIMM → 6 × 128GB RDIMM

Same capacity. Fewer modules, lower cost, room to grow — if the workload profile allows it.

Example · features
Switch w/ MACsec → Same switch, no MACsec

Dropping an expensive feature the customer will never enable keeps the quote sharp and honest.

04 — AI-augmented engineering

Everyone can use AI. The edge is how you orchestrate it.

A single chat window is table stakes. What I actually run is a self-hosted orchestration platform: one orchestrator that plans the work and routes each task across a five-tier model fleet — flat-rate cloud, local on-device, free vision, and two gated pay-per-call tiers — grounded by retrieval over my own notes and gated by human approval on anything metered. Privacy, cost control, and resilience by design.

My home AI platform, mapped onto the inference stack — orchestration down to silicon
01 Orchestration plan · route · gate spend · human-in-the-loop
Hermes orchestrator Discord control plane Approval gate Infisical secrets GitHub
02 API virtual keys · access control · spend tracking
LiteLLM gateway Ollama Cloud API Anthropic API OpenRouter API Gemini API
03 Augmentation & Tools RAG · function calls · web search
Obsidian notes vault pgvector + pgvectorscale Gemini embeddings Tavily web search Tool / function calls
04 Inference Engine serves tokens from each model tier
Ollama (cloud) Osaurus (local server) Anthropic engine DeepSeek via OpenRouter
05 ML Framework the math runtime under each engine
MLX (Apple silicon) PyTorch TensorFlow scikit-learn
06 Model Weights five tiers — the right model for the job
T1 · Ollama Cloud (coding + reasoning) T2 · local MLX 4-bit T3 · Gemini 2.5 Flash T4 · DeepSeek T5 · Claude Sonnet / Opus
07 Compute & Acceleration where the FLOPs actually happen
MacBook GPU (Apple silicon) Home NAS · SSD cache US / EU cloud GPUs Frontier cloud accelerators

Cool at the top, hot at the bottom — orchestration logic down to the silicon that burns the watts.

01

Engineers become PM/PO

The job shifts from writing every line to specifying intent, reviewing output, and owning the outcome.

02

Token cost is the moat

Two engineers reach the same result; the one who got there for a fraction of the tokens stays competitive.

03

Guard against bloat

Code bloat saturates context and quality collapses. I monitor for it and keep the agent honest.

04

Security & MCP validation

In cleared environments, validating every tool and MCP server an agent can reach is part of the job, not an afterthought.

04.1 — Built with Claude Code

Proof, not theory.

These designs were produced by augmenting my own work with Claude Code — the same artifacts that used to take days, iterated in minutes and handed off build-ready.

The takeaway

Depth wins respect. Systems win deals.

I bring both — and I'm building the AI-augmented version of this playbook now. Let's talk about where it fits your team.