Skip to main content
← /writing

topic · 45 essays

Agentic SDLC

An agentic SDLC is a software delivery life cycle in which AI agents do real work at each stage, from planning through operations. People set the intent and verify the results. Most organizations are building one whether they intend to or not. Every tool in the pipeline now ships an agent, and few teams have decided where each one should stop.

My argument across these essays is that the tools are the least interesting part. When turning a clear specification into working code gets cheap, the constraint moves upstream to framing and downstream to verification. The durable asset becomes the spec, kept next to the code and reviewed like code. I call the artifact that carries intent from one stage to the next the Context Spine.

Cheap generation also produces more output than anyone can review by hand. So the checks that repeat belong in gates that run every time, with people writing the standard and deciding whether it's good enough. The measure I use is value per verified outcome, which counts review, rework, and escaped defects alongside the model bill.

I've narrowed the claim since I first made it. A spec can't check its own frame, and in a legacy estate the running system may be the only complete record of what it does. Agents add a security question too. Instructions describe the work and permissions bound it, so the controls have to hold even when a model follows a malicious instruction. Operability belongs in the spec as well, before the 3 a.m. page.

These are field notes more than theory. They come from running coding agents every day on my own apps, and from leading platform and cloud engineering in regulated enterprises. Start with the essays below. If the argument is new to you, the series version takes it in order.

start here · 9 essays

The essays to read first

  1. 01
    The Agentic SDLC: Uniting the AI Tool Sprawl
    16 min read

    Names the system: draws the lanes that turn scattered AI tools into one delivery pipeline.

  2. 02
    The Spec Is the Product
    10 min read

    The thesis beat: code becomes regeneratable output, and the spec becomes the durable asset.

  3. 03
    Spec, Standards, Specialists: How I Actually Build with AI
    25 min read

    The working loop up close: one refined spec, house standards, and a bench of specialist agents.

  4. 04
    Trust the Gate, Not the Actor
    17 min read

    The governance rule: the gate checks the result against the constraints, no matter who produced it.

  5. 05
    The Frame Is the Bottleneck
    10 min read

    As agents get better at execution, the human frame around the work becomes the constraint.

  6. 06
    Two Gates and a Night Shift
    12 min read

    The pipeline up close: three agents, three nested loops, and human judgment placed at two gates.

  7. 07
    Nobody Pages the Agent
    12 min read

    Where the pipeline ends: operability has to enter the spec before the 3 a.m. page.

  8. 08
    The Prompt Is Not the Permission: Securing the Agentic SDLC
    13 min read

    Securing it: permissions bound the work that instructions only describe.

  9. 09
    The Bottleneck Is Never the Stack: What Held and What Moved
    17 min read

    The argument revisited: what held, what I'd narrow, and what I still can't prove.

go deeper

the full list · 45 essays

Every essay on this topic

Engineering Leadership

Over and Over Again

The 10,000-hour rule counted hours, and we now count merged pull requests. Skill comes from feedback at the edge of your ability, and agents take it out of the work unless you design it back in.

15 min read · 2.9k words
★
Agentic SDLC

I Gave the Agents the Merge Button

The lead agent coordinated a weekend of changes across three GSD apps. Shared credentials and incomplete checks showed why every merge needs the same evidence, including mine.

13 min read · 2.5k words
★
Agentic SDLC

Architecture Is What Agents Can't Refactor

Agents made code structure cheap to change. What's left of architecture is what a revert can't undo and what needs someone else's agreement, so that's where platform teams should spend their design time.

13 min read · 2.5k words
★
Agentic SDLC

64 PRs Before Lunch

I merged 64 agent-built PRs across three apps in one morning. The kickoff carried my judgment into the work. A bug I merged past showed me where that judgment still fell short.

15 min read · 2.8k words
★
Platform Engineering

The Agent Is Wearing Your Badge

An agent acting with your credentials can leave a record of your authority without a record of its involvement. Platform teams need to preserve who acted, under whose authority, and how to stop that access.

14 min read · 2.7k words
★
Agentic SDLC

Your Agent Workflow Needs a Release Process

Kiro's frontier engineering principles describe how to work well with agents. Making that practice dependable across a team means treating the shared workflow as a software release, with a version, a scope, an owner, and tests for when it should stop.

10 min read · 2.0k words
★
Agentic SDLC

Worse Is Better, but Simple for Whom?

Coding agents can reduce implementation effort without reducing every obligation that follows. Revisiting Richard Gabriel's argument raises a different question: which design is simplest to verify, use, change, and operate, and for whom?

13 min read · 2.6k words
★
Engineering Leadership

From Faster Code to Better Engineering

Sixteen months of building and writing taught me that AI makes implementation cheaper while raising the value of clear intent, trustworthy platforms, operational understanding, and human judgment.

18 min read · 3.4k words
★
Agentic SDLC

Nobody Pages the Agent

A gate proves a standard ran. It does not prove anyone can operate the result at 3 a.m. Operability has to become part of the spec, and the evidence has to survive the merge.

12 min read · 2.3k words
★
Platform Engineering

The Golden Path Needs an Intent Interface

The golden path made the right way easier to follow. It still asks developers to learn the road. With AI, intent can become the interface while the platform keeps the standards, the controls, and the gate.

7 min read · 1.3k words
★
AI Economics

Where the Saved Hour Goes

AI can create a real time dividend for engineers, and history says unallocated capacity gets absorbed before anyone notices. The case for deciding where the saved hour goes.

13 min read · 2.6k words
★
AI Economics

Everyone Knows You Never Rewrite

For twenty-six years, the full rewrite was the cardinal sin of software engineering. Then one engineer ported half a million lines of Bun from Zig to Rust in eleven days. The taboo was never about code. It was a price tag, and for systems whose intent lives in an executable oracle, the price just collapsed.

19 min read · 3.7k words
★
AI Economics

Paying for Intelligence Twice

Satya Nadella named a real problem: AI workflows generate valuable learning exhaust from proprietary context and corrections. He is right about the asset. My practitioner's edit is simple: own the learning loop, not necessarily the model.

9 min read · 1.7k words
★
Agentic SDLC

Two Gates and a Night Shift

My personal delivery pipeline runs on three Claude agents, three nested loops, and two explicit human approval gates. The gates sit where judgment is cheapest and most consequential, and the real bottleneck moved upstream to the clarity of the ask.

12 min read · 2.3k words
★
AWS

Barometer: I Chased a New Primitive. I Was Standing on It.

I built a serverless internet health monitor because I wanted an excuse to learn Lambda MicroVMs. The better lesson was that Barometer did not need the newest primitive. It needed the boring, correct fit between the tool and the job.

9 min read · 1.7k words
★
Engineering Leadership

Still in the Code

I lead engineering at a Fortune 100 company, and I still ship my own software. Not because leaders need to prove they can code, but because AI is changing the work faster than secondhand models can keep up.

16 min read · 3.0k words
★
AI Economics

The Subsidy, Itemized

Twenty-six days of Claude Code showed $2,556 of API-priced work against a $200 subscription. The lesson was not the total. It was cache behavior, model routing, and a government kill switch that landed in my usage chart. Value lives in verified outcomes, not tokens, and the work has to survive the stop.

10 min read · 2.0k words
★
Agentic SDLC

Trust the Gate, Not the Actor

AI has made generation cheap. The durable advantage now belongs to teams that can encode judgment into gates that verify whether the work should ship.

17 min read · 3.2k words
AI Strategy

The Kill Switch Was Always There

A government order took Anthropic's Fable 5 and Mythos 5 offline. The lesson for enterprise AI is architecture: portability gets you out, verification gets you out safely.

7 min read · 1.4k words
★
Engineering Leadership

The AI Operating Ledger

Enterprise AI is entering its accounting phase. The winners will measure verified outcomes, not just tokens, subscriptions, or generated output.

11 min read · 2.1k words
★
GenAI

The Subsidy and the Severance

AI's free-lunch phase is ending, and that makes AI-driven layoff math harder to defend. Two repricings, one ledger.

13 min read · 2.5k words
AI Strategy

Workflow.ts: When the Code Runs the Agent, Not the Other Way Around

Agent loops let the model pick the next step. Workflows invert that. Code owns the control flow; the model owns the judgment inside each step. Here is the TypeScript file I am running today, type-checked against the live SDK, and the honest answer to whether you should build this now or wait for the official tool.

19 min read · 3.6k words
★
AI Strategy

The Frame Is the Bottleneck

My thesis is that as agents get better at execution, the primary constraint for organizations shifts from technical production to the human-led framing of problems.

10 min read · 2.0k words
★
AI Strategy

Code Review Your Prompts

Viral prompt threads borrow the language of science without the rigor. Here's a four-question code review for any prompt, plus a worked example that shows the gap between sounding authoritative and being right.

14 min read · 2.6k words
AI Strategy

The Agentic SDLC: Uniting the AI Tool Sprawl

Every tool in your product development life cycle is now an AI agent trying to do everything. Here is how to stop the chaos, draw the right boundaries, and build an orchestrated pipeline that actually works.

16 min read · 3.1k words
★
GenAI

The Dark Factory Model for AI-Driven Software Development

Most engineers treat AI-generated code like work from a junior developer they don't trust. Simon Willison gave me a better mental model: the dark factory. Here is what it means, why experience is the raw material, and how to build a system that runs.

5 min read · 999 words