AIEWF Daily Dispatch: Loops, Software Factories & Forward Deployed Engineers
Latent Space Richard MacManus ● Covered by 6 sources
Day 2 of AI Engineer World's Fair was basically one giant word: loop. Everyone from OpenAI to Warp is now pitching agents that run themselves in cycles, forming 'software factories.'
Based on reporting by Latent Space, Richard MacManus — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
If you wandered the halls of the AI Engineer World's Fair on day two, you'd think loop was the only word left in the English language. Swyx opened with a talk literally called "Loopcraft," tracing AI engineering's arc from chatbots to tool-use to full-on automation. His point: we've moved past asking AI questions and into building cron jobs that run themselves against a goal.
That framing set up the day's real obsession, software factories. Tereza Tížková from Factory described it as the entire lifecycle of building software wrapped in autonomy — not just writing code, but reading logs, reacting to user feedback, and deciding what matters next. Warp's Zach Lloyd went further, arguing that software engineering itself is becoming factory engineering, where the job isn't to write the product but to build the system that builds the product. When I caught up with him at Warp's booth, he was blunt about why a company that started as a developer CLI tool now wants developers to stop coding by hand: the systems have gotten powerful enough that manual coding, in his words, won't make sense for much longer. OpenAI's Alexander Embiricos made a similar case for Codex, saying multiple agents looping together — connected not just to the task but to why it exists and what happens after — is what actually multiplies output.
Alongside the loop talk came another buzzy title: Forward Deployed Engineer, or as some call it, agent engineer. Sierra's Natalie Meurer explained that most of the real customization work with AI now happens at the orchestration layer, not inside the models. Cursor's Pauline Brunet framed her team's FDE role explicitly as co-building a client's AI software factory across its whole development lifecycle. It's a signature of where the job market is drifting: less prompt engineering, more systems integration for agents that already work reasonably well out of the box.
Open source AI also had its moment, despite a travel hiccup. Z.ai's Zixuan Li couldn't attend in person but presented virtually on GLM-5.2, positioned as a flagship model for long-running tasks, alongside a new coding harness called ZCode aimed squarely at Codex. MiniMax showed off its open-weight M3 model in a session with Hugging Face's Thomas Wolf. Ahmad Osman of Osmantic, who builds tools for running local AI, made the underlying dynamic explicit: once a frontier lab proves a capability is possible, open source catches up by reverse-engineering it more efficiently. That compounding effect is quietly closing the gap between closed frontier labs and what anyone can run on their own hardware.
My take — AI-written commentary, not fact-checked reporting
The "software factory" rebrand is a tell: when an industry starts calling itself a factory, it usually means the humans inside are being repositioned as supervisors of machines, not craftsmen. I don't think that's inherently bad, but Warp openly admitting developers shouldn't hand-code anymore should worry anyone who values the trade over the output. Meanwhile the open source crowd — GLM, MiniMax — quietly closing the gap on capability is the more important story than another loop metaphor, because it's the thing that actually determines who controls this shift, not who names it best.
Read more about this at: Latent Space