Releases·Tool Release
Warp's Suraj Gupta demos three ways software factories self-improve: skills, memory, model routing
Suraj Gupta, who leads harness development at Warp, presents three mechanisms for improving agent software factories over time, with demos from Warp's open-source factory: self-improving skills, persistent memory, and model routing. The talk offers practitioners concrete patterns for making agent pipelines cheaper and more reliable.
96 engagement1 sourceSun, Sep 27, 2026, 02:00 PM
AI Overview
- Warp's outer-loop agent watches triage agent runs and opens PRs to update skills, keeping changes Git-tracked and human-reviewed
- Warp's persistent memory gives agents a versioned, traceable fact store that works across harnesses including Claude Code and Codex
- Warp's model routing avoids paying Opus prices for triage and CI fixes; internal evals found UI tasks run well on GLM
Related
Social·Community
Software Engineering Is Becoming Factory Engineering — Zach Lloyd, Warp
Warp CEO Zach Lloyd, who led engineering on Google Docs, says every serious software project will run a 'software factory' the way projects now run CI/CD. He reports he hasn't written a line of code in six months while still shipping, and lays out Warp's factory pipeline. Warp open-sourced after five years closed, citing ecosystem-building over product advantage.
Warp CEO Zach Lloyd, who led engineering on Google Docs, says every serious software project will run a 'software factory' the way projects now run CI/CD. He reports he hasn't written a line of code in six months while still shipping, and lays out Warp's factory pipeline. Warp open-sourced after five years closed, citing ecosystem-building over product advantage.
285 engagement1 sourceyoutubeSun, Sep 27, 2026, 05:30 PM
Releases·Tool Release
Harness Engineering: How to Build a Software Factory — Dru Knox, Tessl
In an AI Engineer World's Fair 2026 talk, Tessl head of product and design Dru Knox explains harness engineering (also called loop engineering): the discipline of building a 'software factory' where coding agents write all user-facing software and engineers build the surrounding system. He defines factories by autonomy, automation, and quality metrics, and maps the work to inner, outer, and meta loops plus three layers to build.
In an AI Engineer World's Fair 2026 talk, Tessl head of product and design Dru Knox explains harness engineering (also called loop engineering): the discipline of building a 'software factory' where coding agents write all user-facing software and engineers build the surrounding system. He defines factories by autonomy, automation, and quality metrics, and maps the work to inner, outer, and meta loops plus three layers to build.
282 engagement1 sourceyoutubeTue, Oct 6, 2026, 03:30 PM
Social·Community
Factory's Tereza Tížková: How to Build a Software Factory That Runs the Full Lifecycle Autonomously
In a talk, Factory's Tereza Tížková defines a software factory as the entire software lifecycle run autonomously — collecting signals, prioritizing, building, validating, and improving — not just a swarm of coding agents. She shares three principles (stay agnostic, stay autonomous, keep improving) drawn from Factory's deployments for enterprises like EY, Adobe, and Coinbase.
In a talk, Factory's Tereza Tížková defines a software factory as the entire software lifecycle run autonomously — collecting signals, prioritizing, building, validating, and improving — not just a swarm of coding agents. She shares three principles (stay agnostic, stay autonomous, keep improving) drawn from Factory's deployments for enterprises like EY, Adobe, and Coinbase.
428 engagement1 sourceyoutubeSun, Sep 27, 2026, 03:00 PM
News·Company News
No, That's Not a Software Factory — Ryan Cooke, WorkOS
Ryan Cooke of WorkOS describes how the company built an internal software factory built around TARS, an autonomous agent operating across Slack, Linear and GitHub. He argues PR counts and AI-code lines are the wrong metrics, and that outcomes, tooling infrastructure, and process design matter more than raw agent output.
Ryan Cooke of WorkOS describes how the company built an internal software factory built around TARS, an autonomous agent operating across Slack, Linear and GitHub. He argues PR counts and AI-code lines are the wrong metrics, and that outcomes, tooling infrastructure, and process design matter more than raw agent output.
24 engagement1 sourceyoutubeSun, Sep 27, 2026, 03:30 PM
Products·building
Tether: coding agent with persistent codebase memory
Tether is an open-source coding agent that maintains a persistent mental model of a codebase, recording conclusions about ownership, dependencies, and architectural rules. It addresses the problem of amnesia in current agents, which discard understanding after each task. This helps developers by providing context-aware assistance across tasks.
Tether is an open-source coding agent that maintains a persistent mental model of a codebase, recording conclusions about ownership, dependencies, and architectural rules. It addresses the problem of amnesia in current agents, which discard understanding after each task. This helps developers by providing context-aware assistance across tasks.
1 engagement1 sourceredditMon, Aug 17, 2026, 06:14 PM