Your AI demo is a lie (and how to make it real) | Arcade’s Alex Salazar
Why constraint is your new superpower in engineering
AI that talks is easy, but AI that acts securely is where everything breaks down. We're joined by Alex Salazar, CEO of Arcade, to confront the massive and often underestimated gap between a flashy AI demo and a production-ready system. Drawing from his team's own pivot from building agents to building the tools that secure them, he explains why a working demo is only 1% of the journey. Alex breaks down the four "demo killers" that cause most agent projects to fail: inconsistency, security flaws, prohibitive costs, and high latency.
Alex reveals the counterintuitive solution his team discovered: the key to making non-deterministic AI reliable is to dial up determinism. Learn why giving an AI a constrained set of intention-based tools - like a calculator or a multiple-choice test - dramatically reduces errors and solves critical security challenges that plague open-ended systems. He explains why you can't just wrap existing APIs and must instead build custom, workflow-centric tools for your agents. This is an essential listen for anyone who wants to build AI that doesn't just talk, but acts securely on behalf of your users.
"The big insight that we came to that led us to Arcade was we dialed up determinism... Instead of letting the large language model just figure stuff out... That's when things started working really well." - Alex Salazar
The Download
The battle for browsers is back like it's 2004 🤺
1. Google dodges Chrome breakup but faces new rules 🤖
In a new twist from the courts, Google avoided having to sell Chrome, but now faces new restrictions. The court decided against a full separation, instead imposing nuanced limits on Google's operations within the browser space. These include sharing data with competitors and restrictions on exclusive agreements with mobile phone makers. This development could be a prelude to AI-native upstarts challenging Chrome's dominance, highlighting the evolving landscape of web browsers.
Read: Google avoids break-up but must share data with rivals
2. Atlassian buys into AI browser innovation 🕸️
Not long after the Google news, Atlassian shook up up the landscape by acquiring The Browser Company to create an AI-powered browser tailored for knowledge workers. This move aims to integrate tools into a seamless workflow, addressing the disconnect that traditional browsers have with professional work. While Atlassian faces the daunting task of changing user behavior, their acquisition strategy focuses on innovation over competition elimination. The tech community is watching closely to see how this will reshape our browsing experience.
Read: Welcoming The Browser Company to Atlassian
3. Coinbase's AI coding claim sparks debate 📈
Coinbase CEO Brian Armstrong stirred controversy (yet again) by claiming 40% of their code is AI-generated, aiming for 50% by October. The obvious counter-argument is that AI-generated doesn't mean high-quality coding, often involving grunt work rather than innovative features. And the verbosity of AI-generated code means that the code measured is in need of refinement. Armstrong's push for broad AI adoption reflects the growing importance of AI tools in development. However, there is something to be said for meaningful AI integration rather than just increasing AI-generated output for the sake of more output.
Read: "~40% of daily code written at Coinbase is AI-generated. I want to get it to >50% by October." on X
I’m hosting an MCP demo that connects AI experimentation to ROI 💰
Investing in AI but struggling to prove impact? Join LinearB’s free 35-minute workshop hosted by me, your host Andrew Zigler! We’ll be looking at automations that alleviate bottlenecks in the entire SDLC, not just code gen. I’ll also be demo’ing our MCP server and showing some best practices for natural language insights on your engineering pipeline. I’m hosting two sessions on September 17th and 18th, so register now to catch me there and receive your AI Productivity Guide for Engineering Leaders.
4. Crushing JIRA tickets isn't real impact 📋
Sean Goedecke's latest article challenges the notion that closing JIRA tickets equals productivity. He argues for focusing on impactful work rather than just smashing tickets, sharing personal experiences of investing in passion projects at work that didn't align with company priorities. Understanding what really drives value has always been crucial for developers, even before AI opened the floodgates.
Read: Crushing JIRA tickets is a party trick, not a path to impact
