TLDR Dev 2026-02-09

Atlassian surveyed engineering teams to see who's actually getting value from AI. Their research found that fragmented tools, context switching, and information silos are still the biggest productivity killers for engineering teams — and that AI alone doesn't fix them.

In this live webinar, you'll get exclusive findings from Tanya Chen, Head of Engineering at Atlassian, on how her team actually implements AI in their workflows. Joined by Product Manager Leader Partha Kamal, they'll walk through:

• How Atlassian is scaling AI across the software development lifecycle
• Real stories from engineering teams
• Practical takeaways for leaders driving AI adoption

Secure your spot for practical strategies to quantify AI ROI

StrongDM has developed a "Software Factory" where AI agents autonomously write and refine code based on specifications and scenarios, removing human involvement in coding and review. One of the main innovations making this possible is its "Digital Twin Universe," a system of behavioral clones of third-party services that allows for high-volume, safe, and cost-effective validation of software scenarios.

Geo joins use spatial predicates and are computationally expensive because they lack a clean join key, leading to quadratic complexity. To overcome this, the Floe database automatically rewrites these queries to use Uber's H3 indexes. H3 indexes partition geographical areas into hierarchical hexagonal cells, converting complex spatial intersections into efficient equi-joins on compact integer keys.

AI is diminishing the programmer's craft, consuming human work, and reducing their role to mere oversight. Despite this, there is a career necessity for professionals to embrace these tools to remain competitive.

A seasoned engineer is uneasy about the industry's rapid adoption of LLM-generated code. He's worried about outsourcing critical thinking, as LLMs are best at generation but not as great at reasoning like a human is.

Martin sees two kinds of AI users emerging: power users who've embraced tools like Claude Code and are flying, and everyone else stuck on ChatGPT or Microsoft's enterprise Copilot — which is so bad that Microsoft itself uses Claude Code internally. The result is small teams with API access and coding agents massively outproducing enterprises trapped behind locked-down IT and legacy software.

You're tracking signups but probably not what happens after. Clerk's new retention report measures how users come back after sign-up: 30-day, 8-week, and 3-month cohorts with industry benchmarks. No analytics setup required.

GitHub Agentic Workflows enable repository automation by running AI coding agents within GitHub Actions. Users define these continuous AI workflows in natural language using Markdown files, which are then compiled into GitHub Actions to automate tasks like code improvements, issue triage, documentation, and analysis. They integrate deeply with GitHub, support multiple AI engines like Copilot, Claude, and Codex, and prioritize safety with sandboxed execution and minimal permissions.

Matchlock is a CLI tool designed to run AI agents within secure, ephemeral microVMs, preventing unrestricted access to your machine. It achieves this by providing network allowlisting, injecting secrets in-flight (so they never enter the VM), and blocking all other connections by default. Agents operate in a full, isolated Linux environment that boots quickly and is disposable, ensuring that even malicious actions don't compromise your credentials or data.

This dev tested Claude on three real React/Next.js bugs ranging from easy to hard. It aced the simple one, half-fixed the medium one (but introduced new issues), and completely failed the tricky one, confidently going in circles with wrong fixes. AI is a great first pass for standard bugs, but knowing when to stop prompting and start thinking is the real skill.

Apple silicon Macs achieve their high performance mainly due to their Efficiency (E) cores, which handle numerous background tasks, rather than solely relying on Performance (P) cores. Immediately after startup, E cores are heavily used for processes like Spotlight indexing and initial backups, leaving the P cores largely free for user applications. This architecture is managed by macOS's Quality of Service (QoS) system, which prioritizes foreground tasks for P cores and restricts background tasks to E cores, even when P cores are idle.

Cloud agents like Devin shift AI coding from local pair programming to organizational delegation. Running on remote infrastructure, these tools allow non-engineers to trigger tasks via Slack or Jira. This model scales capacity asynchronously, moving the primary bottleneck from code generation to automated review.

Platform engineering creates a reverse ivory tower when rigid abstractions distance infrastructure from developer reality.

Agent Slack is a CLI specifically designed to enable AI agents to automate various interactions within Slack.

AI's ability to produce merely "good enough" software (or "slop") might create widespread user apathy, ultimately leading to the demise of quality software craftsmanship.