Skip to content
CRAFTDEXA

Theme guide: SYSTEMS

Infrastructure, data and scale.

Systems fail at their seams: between services, between teams, between what was designed and what runs in production. Scaling is mostly the work of finding the limits before the users do.

What is moving now

Sep 2025

DORA: AI amplifies the system it lands in

The 2025 DORA report, built on a survey of nearly 5,000 technology professionals, links AI use to higher delivery throughput but also to more instability, and finds the gains concentrated in teams with loosely coupled architectures and fast feedback. Its advice is to strengthen the delivery platform before expecting AI to speed it up.

SourceGoogle Cloud: the 2025 DORA report (opens in a new tab)

Apr 2026

Agents need sandboxes and defence in depth

Volume 34 of the Thoughtworks Technology Radar (15 April 2026) called for sandboxed execution and defence in depth around AI agents, alongside warnings about cognitive debt from AI-generated code. Treat an agent's runtime like any other untrusted workload.

SourceThoughtworks Radar vol. 34 release (opens in a new tab)

Summarised from public sources as of November 2026. Dates and details change, so follow the link to the source before relying on one.

A practical checklist

  1. 01

    Find the limit before the users do

    Load-test to failure on purpose, and write down where it broke and what broke first.

  2. 02

    Set service-level objectives

    Pick a few numbers users would notice, agree the target, and let the error budget decide how fast to ship.

  3. 03

    Instrument before you need it

    Traces, metrics and logs tied together, so a question about production can be answered without a deploy.

  4. 04

    Keep coupling loose

    Services that can change and deploy independently are the ones that keep getting faster.

  5. 05

    Plan for failure, then rehearse it

    Timeouts, bounded retries, graceful degradation, and a restore that has actually been tried.

  6. 06

    Own the data model

    Names, ownership and retention for every important table, because moving data is the hardest change later.

Where it shows up in our work

Related capabilitiesTechnical Strategy ·System Architecture ·Cloud Infrastructure ·Code Reviews ·Engineering Workflows

Working on something in systems? Tell us what needs to exist.