DEVOPSdigest

Six in ten developers view workplace AI favorably, yet nearly half trust its output only when they can easily verify the output, according to Stack Overflow's 2026 Developer Survey. Previously, the Developer Survey indicated a trust gap, with AI usage rising three years in row, with trust in its output falling. However, the picture this year's data set paints is a much more nuanced one: adoption has now reached a near saturation point, but developers appear to have adapted.

For most of software engineering's history, writing code was the expensive part. Coding agents changed that almost overnight. But in the large, complex systems behind trading platforms, networks and chip design, the cost hasn't disappeared. It has moved. Undo's new survey of 300 senior engineering leaders building mission-critical software found that teams now spend 42% of their week debugging: 9.7 hours on issues found in development and 7.2 hours on issues customers find in production.

The Linux Foundation, the nonprofit organization enabling mass innovation through open source, announced the operational launch of the BUILD Foundation with initial members ASML, Canva, Datadog, EngFlow, Figma and Spotify. The Foundation gives the Bazel ecosystem the leadership and sustained investment needed to take on challenges no single organization can solve alone. The BUILD Foundation launches at a point of inflection for the software industry's reliance on scalable build tooling.

For most of software engineering's history, writing code was the expensive part. Coding agents changed that almost overnight. But in the large, complex systems behind trading platforms, networks and chip design, the cost hasn't disappeared. It has moved ... Organizations can see model usage, token consumption, costs, and other activity. What is often missing is the connection between that activity and the work the business truly cares about.

For most of software engineering's history, writing code was the expensive part. Coding agents changed that almost overnight. But in the large, complex systems behind trading platforms, networks and chip design, the cost hasn't disappeared. It has moved ... Organizations can see model usage, token consumption, costs, and other activity. What is often missing is the connection between that activity and the work the business truly cares about.

I run engineering at a company where AI agents now work alongside my team every day. And the question I keep coming back to is: what did the AI actually contribute? McKinsey found that 62% of organizations are already experimenting with AI agents, while only 39% report any enterprise-level EBIT impact from AI. That gap isn't for lack of data — most AI tools generate plenty of it. Organizations can see model usage, token consumption, costs, and other activity.

Harness has acquired select Augment Code assets, including Cosmos and the Auggie CLI products, the Code Context Engine, and related technology. The team behind these products will join Harness, accelerating its work to make the entire software development lifecycle increasingly autonomous. Cosmos will become Harness Cosmos Software Factory Agent, with a clear role: automate engineering work from idea to code.

Enterprise AI conversations tend to gravitate toward models. Teams compare reasoning capabilities, context windows, benchmarks, and the latest improvements in large language models. When an application produces unreliable answers, upgrading or replacing the model can seem like the obvious response. In production environments, however, the model is often working with a more fundamental constraint: the quality and structure of the information available to it.

In Part 1, I described replacing an SSH-based test execution step in CI with an event published to EventBridge, buffered through SQS, and consumed by a Lambda that dispatches the run to a Fargate task. That's the tidy version. This is the untidy version: the properties of the design that will bite you if you don't plan for them, and what the pipeline's permissions look like when it is done. SQS standard queues deliver at least once, and the EventBridge-to-SQS hop is at-least-once too.

Testlio announced the launch of LeoSuccess™, an interactive AI copilot that orchestrates end-to-end QA execution. LeoSuccess joins LeoMatch™ and LeoInsights™ as the latest addition to Testlio's LeoCore™ platform. It brings the platform's full capabilities into one interface, where teams can manage their crowdsourced testing programs by asking for what they need in plain language. LeoSuccess automates coordination, so coverage is no longer limited by internal bandwidth and capacity.