The mental model of DevOps I had built over the years – two groups, a delivery pipeline, feedback loops from production back to the project, embedding business knowledge into operations – started shifting when we moved to consuming dozens of external services. At our small company building interactive TV apps, we chose to focus on our core business rather than building everything ourselves. Gmail, Slack, CDNs, analytics, mobile services, community platforms – almost no servers of our own.
This worked well until the services started failing in unexpected ways. Coffee breaks upstream meant our builds stalled. Undocumented changes broke pipelines. DynamoDB gave inconsistent behavior across machines. Amazon ELB auto-scaling was too slow for our TV show spikes – we literally had to call Amazon to pre-warm capacity. The maintenance burden dropped, but the risk from external dependencies shot up.
Functions as a service (serverless) pushed this even further. We used Lambda to generate animated GIFs and custom images at scale for kids TV shows – peanuts to scale to 30,000 users for a one-minute interaction. But I turned to promise theory to make sense of it all. A promise is not a contract. It is verifiable, clearly documented, and mutually agreed upon. You cannot make promises on behalf of someone else. If Amazon goes down in Sydney, that is still your responsibility to your customers.
The dangerous pattern is the super agent – relying on one massive provider who becomes your single point of failure. They evolve slower, version changes force you to dance along, and scaling their promise gets harder as they grow. The emerging practices I see are external-facing: status pages, exposing internal metrics, publishing post-mortems, proactive degradation notices, changelogs, giving feedback as gists, direct engineer access via Slack, and even co-designing roadmaps. External services are the next silo to break, and the collaboration and openness we built inside companies with DevOps needs to extend outward.
Watch on YouTube – available on the jedi4ever channel
This summary was generated using AI based on the auto-generated transcript.