Suraj LabAmazon software engineer building backend systems and side projects.

Writing Archive

Long-form writing, separated by theme.

Two active tracks: systems with continuity, and the infrastructure required for reliable agent systems.

Essays

10

Tracks

2

Systems With Continuity

4 essays

Essays on memory, governed state, adaptation, and why useful software should compound over time.

01

3 min read

Software Is Shifting From Tools to Systems

The real change is not that models can answer better. It is that useful software now needs loops for observation, correction, and carried state.

02

3 min read

The Demo Works. The Fifth Run Doesn’t.

First-run output is the easiest benchmark. Repeated work is where weak continuity shows up.

03

3 min read

Most AI Agents Don’t Improve Over Time

Repeated work should compound. Most agents still recompute.

04

3 min read

Context Is Not Memory

Memory that affects behavior needs status, provenance, correction, contradiction handling, and revocation.

Reliable Agent Systems

6 essays

Essays on coding agents, verification, orchestration, and the structure required to ship working software.

05

3 min read

Why Most Coding Agents Still Don’t Ship Working Software

Plausible code is local success. Shipping software requires design, verification, runtime checks, and carry-forward knowledge.

06

3 min read

Deterministic Gates Beat Prompt-Only Control

If a rule matters, move it out of the prompt and into the harness.

07

3 min read

The Goal Is Not Multi-Agent. The Goal Is Reliable Software

Extra agents only earn their place when they isolate a recurring failure mode.

08

10 min read

Building a Coding Agent That Actually Ships

A Claude Code plugin for turning vague software ideas into reviewed, tested, committed code.

09

3 min read

What Building This Taught Me About Diagnosis, Not Retries

Repeated failure usually means the system needs a different model of the problem, not another attempt.

10

4 min read

Using coding-agent to Build a Local-First Knowledge Base

The repo tells two separate stories: a markdown-first knowledge base with one shared core, and a coding-agent workflow whose rules and artifacts stayed visible in the tree.