The cheaper it gets, the more it costs


Tech it From Me Johnny Chivers: TECH IT FROM ME
Issue #02
The Take

The cheaper it gets, the more it costs

AI just made building software cheap. Your bill is about to go up.

There's a 160-year-old economics idea called Jevons paradox. When steam engines got more efficient, Britain didn't burn less coal, it burned far more, because efficiency made coal worth using everywhere. Cheaper per unit, higher total spend. That is exactly what's about to happen to software.

AI has made generating code almost free, and the instinct is to assume that means software gets cheaper. It doesn't. When something becomes cheap, you do far more of it: more services, more features, more prototypes that quietly graduate to production, more repos nobody remembers standing up. The cost of a line of code fell off a cliff, and the number of lines you now own went through the roof.

The Jevons paradox: cost per unit falls while total cost to own climbs

And every one of those lines is a liability, not an asset. Someone has to run it, secure it, patch it, pay the cloud bill for it, and understand it at 2am when it breaks. The cost didn't vanish, it moved downstream, from "writing it" to "living with it." That is the bill that's coming.

This is why governance now matters more than generation. Look at the Dogwood and AWS Transform launches below and you can watch AWS quietly pricing in the cost of cheap code. The teams that win won't be the ones generating the most; they'll be the ones who keep a handle on how much they own.

My take: abundance needs discipline. When output gets this cheap, the winning move is to measure what you're producing, be deliberate about how much you actually keep, and put governance in place before the sprawl rather than after it.

I go deeper on the economics, and what to actually do about it, over on the blog: The Cheaper It Gets, The More It Costs.

Seeing this in your own org? Hit reply and tell me where the sprawl is worst.

Worth Knowing - a few AWS launches worth your time, and why
Dogwood: runtime verification for AI agents

AWS open-sourced a governance language that checks what an agent is allowed to do based on its history in a session, not just the current request. Why it matters: this is the governance layer from The Take in a real, usable form. If you have agents running in production, it's how you keep them within policy.

AWS Transform continuous modernization is now GA

Analyse and remediate tech debt across your repos on an ongoing basis, not as a one-off migration. Why it matters: cheap code piles up technical debt faster than ever, and this tackles it continuously instead of in one big, painful migration.

Vector search for Amazon DynamoDB

Store and query vector embeddings alongside your data, no separate vector database to run. Why it matters: if DynamoDB already holds your agent's memory, semantic retrieval now lives in the same place. One less moving part to own (see: The Take).

Web Search on Amazon Bedrock

Foundation models on Bedrock can now pull live web content for grounding, with zero data egress from your environment. Why it matters: real-time answers without shipping your data out. The obvious next building block for grounded agents.

Signal to watch: Kiro Crew

A persistent, multi-agent workspace inside Kiro that runs efforts in parallel and hands work to subagents across repos and days. Why it matters: this is the AI-DLC idea in product form, and it's exactly what I build with in this issue's video.

On the Channel
What Is AI-DLC? Then We Implement It on AWS with Kiro

What Is AI-DLC? Then We Implement It on AWS with Kiro

AI-DLC is the shift I keep coming back to: AI woven through the whole development lifecycle, not just the IDE. I break down what it actually is, then build it for real on AWS with Kiro.

Watch on YouTube →
New on knowza.ai

Introducing Design Challenges

Anyone can spot the right answer in a list of four. The real exam tests whether you can choose it when nobody shows you the options. Design Challenges hand you a real engagement brief with hard constraints: pick the service, configure it, justify it in a line. Then they run a month-end traffic spike through your architecture to see what breaks. It's the closest thing to sitting in the actual seat.

Try a Design Challenge free →
10 practice questions and 1 design challenge. No card.

That's issue two. If the format's working for you, the best thing you can do is forward it to one person who'd like it. See you next time.

Johnny Chivers
Johnny Chivers
AWS Guru & YouTuber · Tech it From Me

You're getting this because you signed up on one of my channels. No hard feelings if it's not for you; the unsubscribe link is just below.

Johnny Chivers

Practical AWS insight for engineers and the leaders who back them. Each issue: one honest take on what's actually changing in cloud and AI, plus a filtered list of launches worth your time.

Read more from Johnny Chivers

Johnny Chivers: TECH IT FROM ME Issue #01 Welcome to the first one. Here's the deal: I'm not going to relay AWS news you can already get from ten other places. Each issue you get one honest take from me, then a short, filtered list of what's actually worth your attention. Five minutes, tops. Let's go. The Take The real AI shift isn't happening in your IDE Everyone's still watching the robot type. That's the least interesting part. I've spent the last few months in rooms with engineering...