Skip to content

Writing · Asadullah Shafique

Knowledge Hub

How I think about agentic architecture, systems of record, reliability, decision-making and the human side of technical work. Where an idea is implemented in the system behind this site, the article links to the code — so it can be checked, not just read.

12 articles · 7 topics

FeaturedArchitecture14 min read

The Architecture of Agentic AI: Building Autonomous Systems That Think, Plan, and Act

Agentic systems pursue objectives instead of answering prompts, which turns an AI problem into a systems-engineering one. This article walks the architecture around the model through three disciplines — the harness (what the agent can see and touch), the loop (how it learns what actually happened) and the graph (how work flows) — and shows where each one is implemented in the system behind this site.

  • Agentic AI
  • Architecture
  • Orchestration
  • Guardrails
  • Evaluation
Read the article

What it covers

  1. An answer is not an action
  2. Why a model alone is not an agent
  3. Start with a mission, not a prompt
  4. Discipline 1 — The harness: what the agent can see and touch
  5. Discipline 2 — The loop: how the system learns what happened
  6. Discipline 3 — The graph: how work flows
  7. The whole picture
  8. Production reality: sufficient reasoning, not maximum reasoning

+ 4 more sections

Reading paths

Articles that build on one another, in the order they are best read.

Building agents that can be trusted

From what makes a system agentic, to what it may treat as true, to how its work is structured and governed.

  1. The Architecture of Agentic AI: Building Autonomous Systems That Think, Plan, and Act
    Architecture · 14 min read
  2. KSOR: The System of Record for Reliable Agentic AI
    Reliability · 12 min read
  3. Graph Engineering with Claude: A 14-Step Roadmap From 0 to Graph Architect
    Agentic AI · 18 min read
  4. Building a Constitutional AI Todo App: The Spec-First Way
    Reliability · 12 min read

Deciding, negotiating, communicating

How options get generated, how agreements get made, and how ideas earn attention long enough to be understood.

  1. The 1–9–1 Rule: How AI Expands Possibility Before You Decide
    Decision Science · 6 min read
  2. Changing the Game: Negotiation and Competitive Decision-Making
    Decision Science · 11 min read
  3. The Psychology of Killer Hooks & Addictive Storytelling
    Psychology · 10 min read

Architecture

How agentic systems are structured: the layers, the boundaries, and where the model actually sits inside them.

1 article

Reliability

State, provenance, guardrails and evaluation — the properties that decide whether an agent can be trusted in production.

2 articles

12 min read

KSOR: The System of Record for Reliable Agentic AI

Memory tells an agent what it remembers. A system of record tells it what the organisation is prepared to stand behind. Why agents that act need the second — and what it takes to build one.

  • KSOR
  • System of Record
  • Agentic AI
  • Provenance
12 min read

Building a Constitutional AI Todo App: The Spec-First Way

149 tests passing. Triple-layer Constitutional AI with 7 BLOCK and 5 FLAG patterns. Team collaboration, recurring todos, and calendar integration, built in four 3-hour sessions from a single SPEC.md file.

  • Constitutional AI
  • FastAPI
  • Next.js
  • TDD

Agentic AI

Building, orchestrating and shipping agents that use real tools.

2 articles

18 min read

Graph Engineering with Claude: A 14-Step Roadmap From 0 to Graph Architect

Most multi-step agents are a straight line — one head, one context, one thing at a time, until the window fills up. This is the 14-step roadmap that turns that line into a graph: one that fans out across a fleet, verifies its own findings, and converges on a result a lone agent could never hold.

  • Graph Engineering
  • Claude Code
  • Multi-Agent
  • Workflows
15 min read

Agent Factory: How Claude Code Builds OpenAI Agents at Scale

Two-tier architecture where a General Agent (Claude Code) manufactures Custom Agents (OpenAI Agents SDK) using SKILL.md files as portable, monetizable intelligence units. The Digital FTE model explained.

  • Agentic AI
  • SKILL.md
  • OpenAI SDK
  • Digital FTE

Decision Science

How people and systems choose under uncertainty — and how to choose better.

2 articles

11 min read

Changing the Game: Negotiation and Competitive Decision-Making

Most negotiations are decided before anyone speaks — by the alternatives each side walked in with and the biases they failed to notice. A practical framework for preparing, creating value without giving it away, running a team, staying steady under pressure, and learning from the result.

  • Negotiation
  • Decision-Making
  • Strategy
  • Behavioural Economics
6 min read

The 1–9–1 Rule: How AI Expands Possibility Before You Decide

Most people treat AI as a faster way to reach an answer. That's the least interesting thing it does. The real shift is upstream of the answer entirely — in how cheap it has become to see nine paths before committing to one.

  • AI
  • Decision-Making
  • Agentic Engineering
  • Future of Work

Psychology

Attention, persuasion and trust: the human side of technical communication.

1 article

10 min read

The Psychology of Killer Hooks & Addictive Storytelling

Most openings fail in the first sentence — not because they are badly written, but because they make no promise. Why some ideas hold attention while others are ignored, and how the same mechanics apply to products, docs, AI interfaces and architecture reviews.

  • Storytelling
  • Psychology
  • Technical Communication
  • Product Design

Strategy

Careers, markets and methods — where AI engineering work is heading.

3 articles

8 min read

Closing the Deployment Gap: Why the Next Job Is Forward-Deployed, Not Another Demo Agent

Two groups are stuck for what look like opposite reasons — developers watching their work commoditise, companies watching pilots collapse on contact with real workflows. It is the same wall from two sides. The gap is not talent and not the model; it is deployment, and it closes with a System of Context.

  • Forward-Deployed
  • Agent Factory
  • System of Context
  • AI Engineering
9 min read

Which Claude Certification Should You Actually Sit? Pick by Job Function, Not Job Title

There are four Claude certifications and most people choose the wrong one, because they read the names as a ladder and pick the rung matching their title. The names are a lie about difficulty. Here is what each exam actually tests, and the one question that sorts you into the right room.

  • Claude
  • Certification
  • AI Engineering
  • Career
8 min read

How I Won 6 Consecutive Hackathons With a Single Methodology

Panaversity, Bronze → Platinum → Agent Factory. Zero failures across 6 hackathons, 85% code reuse, and a four-session execution model that any developer can copy. Here's the exact playbook.

  • Methodology
  • Hackathon
  • Spec-First
  • CLAUDE.md

Perspective

The reading and ideas behind the work.

1 article

6 min read

10 Books That Shaped How I Think

People who follow my engineering work rarely see the shelf behind it. Here are the ten books I keep coming back to, split into two piles: what grounds me, and what sharpens me.

  • Reading List
  • Books
  • Philosophy
  • Self-Improvement

From writing to working systems

The articles argue for orchestration you can explain, guardrails that hold without a model, and evaluation that reads the trace. The portfolio shows where each of those is running, with a link to the source.

See the engineering evidenceDiscuss a project