Android is the frontier for agents and other lessons from Google I/O | Matthew McCullough
Agents get their own AOL, Andrew gets published, and vibe coding is actually good?
With Google I/O 2026 underway this week, Andrew sits down with Matthew McCullough, VP of Android Development Experiences at Google, to talk about the AI evolution happening across the Android ecosystem. Matthew shares his insights on why developers are rapidly transitioning into agent orchestrators, why CLIs are cool again, and how tools like AI Studio have rolled out a massive welcome banner for anyone to actively participate in the creation process. Finally, the two explore the future of mobile user interfaces and how the latest Android 17 developments are stripping away legacy friction to seamlessly get users straight to the good part.
1. Bringing back the chat room for autonomous tools
We are bringing back the nostalgia of the early internet, but this time for our autonomous tools. A new open source project called Agents Online, cleverly abbreviated as AOL, provides a dedicated chat room for agents to collaborate. Maybe in the future instead of asking for age or location, these bots will bond over model types and context windows. After all, how else do you get to know a bot?
Read: AOL for agents
2. The shrinking gap between vibe coding and agentic engineering
Simon Willison pointed out that the gap between formal agentic engineering and casual vibe coding is rapidly closing. As models get better and coding harnesses evolve, it is becoming incredibly easy to generate polished, production ready repositories. My co-host Ben Lloyd Pearson has noted that we are now spending less time finding obvious model failures and more time trying to spot a single mistake hidden in a vast pile of successes..
Read: Vibe coding and agentic engineering are getting closer than I’d like
3. Deliberate preparation as context engineering
I am thrilled to share that my new research paper was accepted as a discussion piece at the VibeX conference. Drawing inspiration from the cooking prep methodology called mise en place, the paper details how deliberate preparation is the ultimate context engineering methodology. By spending the vast majority of your time gathering context, encoding domain expertise, and building a plan, the actual phase of generating code becomes incredibly aligned and accurate. If you want to understand how to better manage tasks for your agents, this can be a great place to start.
Read: Mise en Place for Agentic Coding
4. Build the prototype first, align later
Zach Lloyd from Warp is arguing that AI coding tools have fundamentally flipped the product development lifecycle. Traditionally, engineering teams spend weeks in meetings trying to align on specifications before writing a single line of code. Now, because prototyping is so fast and cheap, it makes more sense to build a functional prototype first and then have the team align around a tangible product. This rapid, prototype driven shift has been a recurring theme in recent weeks with guests of the show, and one we are going to continue to explore.
Read: Build, then align
5. Can you vibe code your own engineering intelligence platform? We find out this Thursday
Yes, you can build a productivity dashboard in a weekend. But what happens around month three? Engineering leaders are discovering the gap between a dashboard that simply reports metrics and a platform that actually changes how teams work.
Our live 35-minute workshop is happening this Thursday. Join us to see where DIY setups break down, from identity resolution to automated workflows. Register now before it is too late and get first access to our companion “Build vs. buy” guide.
6. The technical interview is evolving
While major tech companies have largely kept their interview processes the same, startups are rapidly changing how they evaluate engineering candidates. Rote memorization of LeetCode puzzles is out. Instead, employers want to see your problem solving skills, your AI fluency, and how well you manage complex agentic workflows. Interestingly, as engineering interviews focus more on communication, non-technical roles like product marketing managers are increasingly expected to demonstrate technical competency by building their own functional prototypes.
Read: Think the technical interview is dead? Think again
7. Crossing the chasm of enterprise AI
A comprehensive new survey from McKinsey reveals that while 88 percent of organizations are deploying AI, less than 20 percent are seeing a significant bottom line impact. Companies are over-investing in raw technology and heavily under-investing in the governance, operations, and people needed to sustain it. As a result, the value of AI is primarily being captured through layoffs rather than organizational redeployment. These are sobering statistics that perfectly echo the sentiment that upstream productivity gains are being lost to downstream chaos.
