Agents moved where the work happens (and using MCP to find it again) | Slack’s Jaime DeLanghe
Empathetic leadership for tech overlords, a good backlog completes itself, and who’s agent is this, anyways?
This week on Dev Interrupted, Slack’s Chief Product Officer, Jaime DeLanghe, joins the show to explain why enterprise AI value depends on embedding custom bots directly into your existing team communication loops rather than deploying them inside isolated, single-player chat silos. She breaks down the platform's shift toward open ecosystem standards like the Model Context Protocol (MCP) and how dynamic UI frameworks are transforming standard channels into active execution environments. Jaime details the operational realities of managing autonomous software fleets, including a striking look at how leading companies are placing hundreds of custom agents directly onto their corporate org charts.
1. The final day for Fable on your subscription
Fable suddenly returned to our lives last week, but is already on the way back out the door again! Yes, today is the last day Fable is available on a subscription plan, then it is API pricing starting here on out. And don’t worry: this week’s special guest Kelly Vaughn verified that it does indeed know there are three R’s in strawberry, but the real takeaway is its internal model routing. It relies on models like Opus and Sonnet underneath, pointing toward a future where we stop seeing massive foundational drops and start seeing highly specialized, domain specific intelligence that scaffolds through routing.
Read: Redeploying Fable 5
2. Keeping human context in the loop
HumanLayer co-founder and Dev Interrupted alum Dex Horthy took the stage at the AI Engineers World Fair this week to discuss a growing problem in agentic engineering. Teams are pushing fast code throughput with poorly trained models, only to have critical flaws surface months later because they lack auditability. Optimizing your harness is great, but putting the human back in the cognitive loop is the actual differentiator for delivering durable software. We have to ensure human context remains at the center of the conversation when incidents eventually arise.
Read: It’s Time To Put Humans Back In The Software
3. Solving the agent identity crisis
As agents roam around your codebase, establishing who did what and under whose authority is becoming a massive security concern. This technical deep dive outlines three approaches to agent identity, ranging from borrowing user credentials to building complex, durable agent IDs. Kelly smartly pointed out that this governance should be treated like a stepladder. Start small by borrowing credentials and rely on baseline instrumentation before you accidentally over-engineer a sprawling identity architecture that your team cannot support.
Read: Three Ways to Give an AI Agent an Identity
4. Testing agents against real enterprise data
Friend of the show Bryan Bischof hosted the America’s Next Top Modeler hackathon last year, dropping a messy, real-world enterprise dataset onto 150 engineers (including moi) to see how they would solve complex queries. The resulting breakdown reveals exactly where agents stumble when attempting data science, specifically when sequencing operations and data joins. If you throw a vague Jira ticket at an agent, you are just wish casting. You have to translate those vague goals into precise, step-by-step instructions if you want your AI workflows to succeed.
Read: Benchmarking AI Agents for Real Data Science
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. Ending arguments with AI prototypes
Disagreements are a fundamental part of building software, but arguing for correctness often forces people to dig their heels in further. You no longer have to argue about roadmap priorities or technical directions. Because, if you know your path is correct, then it’s never been easier to use AI to show the art of the possible. Data drives the agreement, and providing a working example is the best way to prove your point without engaging in a fruitless contest of being right.
Read: Why I Stopped Arguing With People
7. Avoiding the tech leader survivorship bias
Charity Majors launched a new advice column on Stack Overflow, and her latest piece tackles the survivorship bias that creates supervillain-esque tech leaders. Our industry has a bad habit of putting successful CEOs on a pedestal while ignoring the teams and circumstances that actually made their achievements possible. Charity offers incredibly empathetic advice for navigating this reality, reminding us that you can be a highly effective, strategic business operator without losing your humanity in the process.
Read: Paging Charity! How can engineering leaders avoid becoming Bond villains?
8. A new era for the engineering backlog
Kelly recently wrote an excellent piece exploring how multi-threaded agentic engineering is changing our relationship with the backlog. For the first time, we actually have the capacity to clear out technical debt and paper cuts, provided we structure our work correctly. Keeping the lights on can be a compute cost, not a time cost, if you have the right mechanics to capture ideas in actionable formats. To take it to practice, she suggests a two-lane system where engineers focus on one big rock task while running side channels for micro-reviews of agent outputs. She points out that since developers cannot magically multiply their own productivity without assistance, investing in durable CI tooling is more critical than ever.
Read: The backlog is finally getting its moment
