Martin Trojer’s talk was the one whose title had caught my eye in the XT26 line-up, and I’m glad I got to sit in for it, because Martin’s session was one of the least doom-y takes on AI and dev work I’ve heard in a while.
For quick context: a year ago Martin (Meta) was telling people to lean into coding agents, much to the continued dismissal of others. Now, with a year having passed, he’s very much looking vindicated. Giving his honest answer to the question “Where is this all actually going?”, he says nobody really knows yet, we’re still in the middle of that storm, but there’s an undeniable truth that “the Rubicon has been crossed, and the Genie will not go back in the bottle.”
What I found fascinating was how Martin framed his current use of agentic tooling as a ladder. AI hands you a bucket full of “plausible code”; tests, comments, looks the part, secretly weighing you down, causing you to fall over the second you lean on/try to progress with it. Every rung up you travel with the bucket is human effort: reviewing, ironing it out, until eventually your bucket, now much lighter, only contains proven code that’s good for production, and you’re able to stand, raised up, having scaled the ladder. The trap is skipping the climb, accepting plausible code blindly, and racking up cognitive debt - shipping fast while understanding diminishes, until nobody remembers why or how anything was actually decided upon.
Another piece of practical advice that landed for me, almost as a corollary to the ladder & bucket, is that devs need to “release their inner hoarder”. In a world where you’re leaning into agents to go faster and faster, these old packrat instincts (saving snippets, recipes, hard-won lessons, the fix that took you a whole afternoon to land), suddenly start to not only earn their keep, but become crucial to crafting the perfect context needed to accelerate agentic capabilities. Gone are the days of a well-curated hoard just being notes-to-self; it’s substance you can feed straight into your agent. Fuel it with your practically proven patterns, and now the initial bucket is starting further up the ladder. Your hoard will only make the AI better, meaning less climbing for you, meaning more productivity codewide.
His framing for what devs are actually for now really sticks out to me: agents don’t feel the pain of code, YOU as a developer do. That intuition of “wait, why is this like this”, the ache of seeing code that’s completely wrong, is what’s worth protecting and cultivating - it’s what will drag you into truly understanding the system instead of just nodding along.
He ended with an excellent quote: “you can outsource your thinking, but you can’t outsource your understanding.”
I walked out more optimistic than I went in. Personal taste, judgement, experience, and a willingness to say no will be going up in value, not down.
Download Martin’s slides here.
Want to be the first to know when the XT26 talks drop? Register here and we’ll send the link straight to your inbox.
To discover our Financial Services offerings, visit our page.
