How LinearB helps Kraken finds hidden bottlenecks across thousands of engineers | Nik Sudan
The discernment horizon, loop-driven development, and a wizard’s very defensible pond
Are you confusing a skyrocketing AI token bill with actual engineering value? This week on Dev Interrupted, Kraken's Engineering Operations Lead, Nik Sudan, joins the show to break down the harsh realities of moving agentic AI projects from pilot to production without compromising code health. He unpacks why raw AI adoption is a flawed vanity metric, detailing how his team uses tools like the LinearB MCP server to combine high-level engineering metrics with granular repository data to uncover hidden workflow bottlenecks. Finally, Nik reveals his exact playbook for translating complex data, like P90 cycle times, into a clear, business-driven narrative that secures vital buy-in from non-technical stakeholders.
1. Are we approaching the discernment horizon?
Steve Yegge just dropped his latest missive on the state of agentic engineering. He argues that the exponential capability curve of AI might be flattening out, not because the models are stalling, but because governments are locking down access to frontier models like Fable. As we get pushed toward the discernment horizon, the ceiling on these systems becomes human based, where we can no longer evaluate or understand the outputs. My co-host Ben Lloyd Pearson aptly noted that a plateau might actually give teams a firmer footing, allowing us to establish best practices that will not just change again in another couple of months.
Read: The Flat Curve Society
2. A magical allegory for tech disruption
Scott Werner is an absolute genius with this fun allegory. His latest essay is a brilliant, allegorical story about a wizard, a traveling sorcerer, and a very defensible pond that perfectly captures the anxiety of established companies facing AI disruption. It is a relatable, children’s book style narrative that makes the current reality of our industry immediately obvious. There are limits to what AI can replace, and this essay is a fantastic reminder of where those boundaries lie.
Read: The Wizard With the Very Defensible Pond
3. Evolving beyond the CI/CD pipeline
As we level up our AI maturity, our coding experience needs to evolve. We are moving past the days of simple autocomplete and prompt engineering into a world of loop-driven development. We dive into the latest loop engineering whitepaper from Anthropic, which focuses on managing the weekly or monthly cycles of how agents learn, deliver assets, and interact with human domain experts. If you have an automated factory running in the background, your next challenge is designing the exact rules for how those agents bring their work to your desk for review.
Read: From Test-Driven to Loop-Driven Development
4. Why model divergence is actually a feature
We often test different models by giving them the same prompt and comparing the results like a race. However, this article argues that when highly capable models produce wildly divergent answers, it says more about your own unresolved thinking than the tools themselves. Every model has a unique linguistic bias. When they disagree on framing, it is usually a sign that your initial prompt was too ambiguous. Do not just pick a daily driver out of convenience or declare a “winner”. Use those different perspectives to test your assumptions and refine your ideas.
Read: A failed universal language explains why you keep picking the wrong AI output
5. Life beyond tokenmaxxing
Stop relying on code volume metrics. If you missed our live workshop, you can now watch Ben and I break down how to measure AI’s real impact across the SDLC on demand at linearb.io. You will get the exact operational model you need to answer board-level ROI questions, plus access to our new APEX framework on measuring AI efficiency.
6. Structuring knowledge with plain text
Organizing your domain expertise into a durable store is a baseline requirement for modern knowledge workers. Google’s Open Knowledge Format is a lightweight spec that uses interlinked markdown files to give your agents readable, structured context without complex databases. While it’s certainly funny to watch people re-invent what is basically YAML over and over again, Ben and I both agree that compiling your knowledge into plain text is incredibly effective for agentic systems. Don’t be sleeping on your second brain. Just wait until we all start realizing that semantic HTML5 might actually be the better long-term solution (are we ready for this conversation yet?)
Read: Building a Local Knowledge Base in Google’s Open Knowledge Format (OKF)
7. Midjourney’s leap into medical imaging
Midjourney is taking AI into the physical world, announcing a new spa experience that uses dolphin-like echolocation to scan your body in a shallow pool of water. While it sounds like a joke at first, this is exactly where AI needs to be applied next. We are moving beyond just model advancements and focusing on real-world applications. If they can pull off these bold claims about accuracy and speed, this could be a massive leap forward for medical imaging accessibility.
Read: A New Era of Midjourney
