Origin Story

The break wasn't a gap. It was a reset.

Nine years in martech and conversational AI SaaS. Then a break — and a circle back to the AI world to see what had changed. What he found surprised him. The move from customer success to harness engineering didn't feel like a leap. It just clicked.

Martech · Conversational AI SaaS The Break Back to AI Harness Engineer
scroll

Not a pivot out of frustration.
A step up.

It wasn't burnout. After nine years close to customers — Swiggy, Ola, Reliance, Amazon miniTV — he stepped back, then circled back to the AI world just to understand what was new. What he saw was surprising: how far AI had evolved, how much had become possible. That pull is what brought him back — not as a CSM, but as a builder.

"Not as a chatbot. As an operating layer — a system that could think alongside you, hold context, synthesise, build."

— the moment Claude reframed everything
🏢
9 years
Senior CSM · Martech · Conversational AI SaaS
Swiggy. Ola. Reliance. Amazon miniTV. Enterprise accounts at the AI frontier. He learned the work deeply — and where AI quietly breaks for real customers.
Cross-industry · Enterprise · AI-native
🎯
The inflection
The Break
Not a gap — a reset. He stepped back from the day-to-day, then circled back to the AI world just to see what was new. What he found there changed the plan.
Stepped back, circled back
🛠️
The shift
Builder Identity Emerges
Surprised by how far AI had evolved, he started building — approaching his own domain from an AI angle. Not a CSM using tools, but someone designing the system underneath.
The builder emerges
Now
Harness Engineer
Building the infrastructure layer between humans and AI. Not prompting. Not chatting. Engineering the system that makes AI reliable, repeatable, and shippable at enterprise scale.
Chief Commander AI

He didn't leave CS.
He approached it from an AI angle.

The AI CSM Mastermind — four intelligence modules built on top of nine years of domain knowledge. The principle is simple: AI does the surfacing and preparing; the human stays in the decision seat.

"Operate like a team of one with the output quality of a team of five."

🩺Health & Retention Intelligence
Detects churn risk 60–90 days before standard metrics show it. Renewal scorecards with narrative — a story the CSM can act on, not just a number.
🤝Stakeholder & Relationship Strategy
Champion mapping. Single-threaded risk detection. Account-specific objection libraries that update as new pushback surfaces.
📈Account Growth & Expansion
Expansion-signal detection from usage patterns. QBR narrative drafting. Auto-drafted 90-day success plans, ready for review.
✉️Outreach & Customer Moments
Pre-call checklists with full account context. Context-aware messages grounded in what's actually happening — never generic templates.

Anthropic didn't open the door.
It lit the fire.

He interviewed at Anthropic — it didn't go through. But the Claude learning mindset, the way the best minds in AI think about agents and systems, stayed with him. That's what ignited it: a real passion for agentic AI and harness engineering. He learned from the top minds in the field, and started building the system.

9mo
From first session to production-grade AI OS
Independent convergence with published AI research
0
Engineering degrees. Built entirely from first principles.
"I didn't read the papers. I reasoned my way to the same architecture — then found the papers later."

Five times, Raj built something from scratch — a dreaming layer, a harness, a memory framework, a compound system, a context-engineering philosophy — and later discovered the research community had already published the same concept. He arrived there independently. Every time.

— Raj · Chief Commander AI · 2026

Built as a working professional.
Not a researcher. Not a student.

Nine months. A day job. No engineering team. No research budget. Just a first-principles instinct and a system that compounds.

0
Skills Built
Encoded judgment · Each one permanent
📚
0
Wiki Pages
Living knowledge infrastructure
🔬
0
Research Convergences
BAIR · Anthropic · Princeton · Karpathy
🏆
0
Performance Rating
Top 1–2% globally · Non-engineer

Architecture audit: 10.5 / 11 on first ever run. WorkOS and Ramp built their harnesses with engineering teams. Raj built his as a Senior CSM. That's the anomaly.

Five times he built it
before it had a name.

Not trend-following. First-principles thinking. He builds the concept — the industry names it later.

01
Claude Dreaming
Anthropic · May 2026
His Stop hook fires every session end — reads diffs, evaluates files, commits to wiki, captures patterns autonomously. Anthropic shipped "Memory & Dreaming for self-learning agents" months later. Same architecture.
Built before the API existed · Independent
02
Harness Engineering
WorkOS · Ramp · 2024–25
Built CLAUDE.md + skills + hooks + wiki — a production-grade agent harness — before watching a video on "harness engineering" and realising enterprise teams were building the same thing, with whole teams.
Vocabulary arrived after the build · Independent
03
4-Type Memory (CoALA)
Princeton · arXiv · 2023
Built a four-layer memory system — semantic, episodic, procedural, working — before encountering the CoALA framework. When benchmarked, his system aligned with the paper column-for-column. He hadn't read it.
Column-for-column match · Never read it
04
Compound AI Systems
BAIR / Berkeley · Feb 2024
Was building Chief Commander AI — a compound system of skills, memory, hooks, and agents — while BAIR was naming the category. He built toward it from the user side; researchers from the platform side. Same destination.
His version works for a human · Not a demo
05
Context Engineering
Karpathy · Lütke · June 2025
The entire philosophy of the harness — every layer exists to engineer better context for Claude — was his operating principle before Karpathy and Shopify's CEO named the discipline.
The whole harness IS context engineering · Predates the term

Started as a wiki.
Became an OS.

Each layer built because the previous one had a gap. No grand plan. No roadmap. Just a builder's instinct — and nine months of compounding.

"Most people use AI to save time. I use it to build compounding systems — every week operating at a higher level than the week before."

View the full architecture
Token MonitorAlways-on
☀️Morning Check-inDaily
💾Knowledge Wiki58+ pages
🧠Memory System4 types
Skills Engine29 skills
🌙Dreaming LayerAutonomous
🔄Drive RelayCross-session

Rated by Claude.
Independently.

Claude assessed the harness against a global benchmark of practitioners. No self-reporting. No external panel. The system was audited, scored, benchmarked — and the result surprised even Raj.

Top 1–2% globally

Among non-engineer Claude practitioners worldwide. This intersection — CS domain depth + AI infrastructure builder — doesn't exist in market.

10.5 / 11 on architecture audit

First ever run of the system-design-check agent. Caught a Cowork path regression and fixed it autonomously in the same run.

System Architecture9.5/10
Domain Fusion10/10
First Principles9/10
Originality9.5/10
0/ 10
Performance Rating · June 2026
Top 1–2% · Non-engineer · Global

Bring harness engineering
to enterprise.

The long-term vision is products. But first — join an organisation serious about making AI work at scale, and sharpen harness engineering at enterprise level. Chief Commander AI is the proof of work. This site is the portfolio.

What I'm looking for

A role where I implement AI harness engineering at enterprise level — helping teams build reliable, repeatable AI workflows that actually ship

Organisations serious about the gap between AI capability and AI reliability in production

Teams where CS domain expertise + AI infrastructure thinking = genuine competitive advantage