PLATFORM ENGINEERING NEWSLETTER

Hi there,

​

Welcome to this fortnight's roundup. A theme kept popping up in my reading: the teams getting real value from AI are the ones treating it as a platform problem.

 

Walmart's LLM playbook makes that case with hard numbers, and the Harness Unscripted panel in London landed in a similar place on guardrails and human review. Closer to home, Derik has written up how we rebuilt our Backstage integration so that Kratix Promises feel ready to use in the portal, rather than like raw YAML.

​

There's also fresh survey data from Octopus, a great QCon talk from Matthew Liste on "ice cubes, not snowflakes", and a few podcasts for your commute.

​

Quick Brain snacks 🧠

• From Raw YAML to Ready-to-Use: Rebuilding Our Backstage Integration for Platform Adoption
  Derik walks through how we redesigned the Backstage templates and catalog entities that SKE generates from Kratix Promises, with clearer request forms, Backstage Groups instead of namespaces, and live health and provisioning status. A good read if your portal is the front door that most of your developers actually use.

• The Platform Engineering Playbook for Production LLMs
  Aditya Mulik shows how a retail LLM system went from a 15% hallucination rate to 1.5% without touching the model, simply by running the LLM stack as shared platform infrastructure. Prompt registries, intent gates, per-request cost attribution, and default-deny MCP tool authorisation are all features.

• The Future of Platform Engineering report
  Octopus surveyed 379 platform practitioners and found that unclear strategy hurts more than technical complexity. The finding that mandatory adoption correlated with hitting goals is sure to spark some debate in platform team retros.

​

Fresh Off the Press 🗞️

• Harness Unscripted: AI agents may be cool, but firms aren't taking humans out of the loop
  Engineering leaders from Aviva, Deutsche Börse, Anthropic, and AWS on tiering automation by risk and keeping humans in the review loop. I especially liked the call for platform engineers to codify regulatory expert knowledge so agents can use it.

• Building a Collaborative Platform Culture in a Bank
  Ben Linders covers a KubeCon EU talk arguing that a platform is a collaboration system first and infrastructure second. The "Genius Bar" sessions and super user awards are practical ideas you could try next sprint.

• How to justify investment in a platform
  A useful framing for when you need to make the case upwards: most organisations already run an unfunded, informal platform whose costs nobody tracks. The piece offers tips for measuring those hidden costs and weighing the long-term maintenance burden of build versus buy.

​

Click. Watch. Enjoy. 📹

• Building Resilient Platforms: Insights from 20+ Years in Mission-Critical Infrastructure
  Matthew Liste's QCon London talk is packed with hard-won lessons from financial services. "You can only vend ice cubes, not snowflakes" is a line I'll be borrowing for a while.

• Building GenAI Platform at DoorDash
  Siddharth Kodwani and Swaroop Chitlur explain how DoorDash built an LLM gateway and agent gateway now serving more than 5,000 internal users. Their shift from vendor-first to open-weights models, while keeping portability, is well worth hearing.

​

Plug In and Tune Out 🎧

• How Meta reduced diff authoring time by 40%
  On the Engineering Enablement podcast, Meta researcher Moritz Beller shares how they measured and improved the time engineers spend authoring code changes. Solid evidence-led thinking on developer productivity.

• DOP 372: Using AI Every Day Doesn't Make You Good at It
  Darin and Viktor talk with Chris Daigle about why daily AI use doesn't automatically build real skill. A healthy reality check for anyone rolling out AI tooling across their engineering org.

• DOP 371: Should You Vibe Code Your Own CRM?
  A fun build-versus-buy debate for the AI era. The same questions apply to the internal tooling platform teams are tempted to vibe code.

​

​

Until the next time,

 

Daniel