Systems | Development | Analytics | API | Testing

During this week’s AWS us-east-1 outage, Ably maintained full service continuity with no customer impact. This was our multi-region architecture working exactly as designed; error rates were negligibly low and unchanged throughout. Any additional round trip latency was limited to 12ms, which is below the typical variance in any client-to-endpoint connection, and well below our 40–50ms global median; this is imperceptible to users and below monitoring thresholds.

Vendor lock-in is one of the biggest risks in government IT. For many agencies, it starts subtly. You deploy a commercial off-the-shelf (COTS) procurement system, only to realize that any time you need a change—a new workflow, integration, or feature—you have just one option: pay the vendor to build it or wait for the next product update, which could take years. That’s vendor lock-in. It can also mean limited flexibility in how you manage and support the system.

As a product manager, you want analytics that add value without adding complexity. Automated insights do the heavy lifting inside your app, surfacing key changes for your users without making them dig. The result is happier customers, fewer support tickets, and more time for you to focus on roadmap and innovation.

Migrating from self-managed Apache Kafka connectors to fully managed connectors has been a persistent challenge for data teams working on Confluent Cloud. While Confluent-managed connectors deliver enterprise-grade features, seamless upgrades, and comprehensive support that add up to significant development and operations cost savings, the journey to get there often feels daunting and opaque.

It’s 2 AM after the export workflow failed again. Your users are still waiting for their invoices because your team is in the office manually reformatting 500 of them. By morning, their compliance officer is busy calling when the system times out trying to generate a 100-page report. And before you’ve even had coffee, your biggest customer questions data accuracy because last month’s statement looks nothing like this month’s. The cost?

Real-time data synchronization is critical for businesses that rely on APIs to connect systems like CRMs, billing platforms, and inventory databases. It ensures that updates happen instantly, preventing issues like outdated information, missed opportunities, or operational inefficiencies. Here’s a quick breakdown of how to choose the right tool: Integration: Look for tools with built-in connectors for databases (e.g., SQL Server, MySQL) and cloud platforms (e.g., Snowflake, Amazon RDS).

2025 will be remembered as the year AI wrote poetry, passed medical exams, and stumbled with spectacular blunders. For every impressive breakthrough, there was an equally impressive facepalm.

Some topics are just too trendy to overlook, and AI in testing is definitely one of them. A few weeks ago, we shared a blog post introducing the integration between an MCP server and OctoPerf, highlighting the many benefits it brings. To illustrate this in action, we recently hosted a webinar led by Thomas Pitteman, performance testing expert at Adeo and OctoPerf power user.

In September 2025, we brought AI Engineer to Paris for a two-day conference with 700+ attendees, 25 sponsors, over 47 talks cross 5 tracks. This is the third post in a 4-part series on the event: If you missed AIE Paris, you’ll be able to watch the replays on the Koyeb YouTube channel. Main Stage Day 1 Main Stage Day 2 If you’re not familiar, the team behind the AI Engineer has been running AI Engineer conferences in the US for the past three years.

Replaying traffic is a core tool in your toolbox when you need to reproduce a tricky bug or validate how your app behaves. Traffic replay is especially valuable for testing complex software applications that rely on APIs and microservices, where integration and functionality must be thoroughly validated.