5Latent Space (swyx)·
Introspection co-founder explains autoresearch and self-improving agent loops
Roland Gavrilescu, co-founder of Introspection, discusses the concept of 'autoresearch' — a feedback loop enabling AI agents to iteratively improve themselves — in a Latent Space interview. The conversation covers agent 'recipes,' self-improving loops, and the continued role of humans in what Gavrilescu frames as a software factory paradigm. The piece offers a practitioner-level view of how agentic research pipelines are being designed and operationalized.
Related guides (2)
Related events (8)
AIEWF Daily Dispatch: Autoresearch and the tension between AI and human agency
A conference dispatch from AI Engineer World's Fair 2026 covers debate between proponents of fully automated 'software factory' and 'autoresearch' visions versus speakers defending human understanding and control. The piece captures live tension at a major practitioner conference around how much autonomy AI systems should have in research and software development workflows. The framing surfaces a recurring fault line in the agent-tool ecosystem between automation maximalism and human-in-the-loop approaches.
Andrew Ng outlines three-loop framework for agentic software development
Andrew Ng describes a 'loop engineering' framework for building software with AI coding agents, comprising an agentic coding loop (agent writes/tests/iterates autonomously), a developer feedback loop (human steers at higher product level), and an external feedback loop (user testing, A/B). The piece contextualizes the buzzphrase popularized by Claude Code creator Boris Cherny and OpenClaw creator Peter Steinberger. Ng argues humans retain a 'context advantage' over AI systems that justifies continued human-in-the-loop involvement in product decisions.
Lossy self-improvement
This commentary from Interconnects argues that AI self-improvement is a real phenomenon but that inherent lossiness in the process prevents it from leading to fast takeoff scenarios. The piece appears to engage with the debate over recursive self-improvement and its implications for AI risk timelines. It offers a nuanced middle-ground position: acknowledging self-improvement capability while contesting the discontinuous-growth narrative common in AI safety discourse.
Latent Space: Skill engineering and the case against one-shot AI design
Paul Bakaus discusses 'skill engineering' as a design philosophy for AI-assisted workflows, arguing against fully automated one-shot AI pipelines in favor of keeping humans in the loop. The conversation centers on Impeccable, a tool or approach Bakaus is developing, and the concept of 'loopmaxxing' — iterative human-agent collaboration cycles. The piece addresses why current agents still require human steering to produce high-quality outputs.
OpenAI Introduces Deep Research Agent
OpenAI has launched 'deep research,' an agentic capability that uses reasoning to synthesize large volumes of online information and complete multi-step research tasks autonomously. The feature is initially available to ChatGPT Pro users, with rollout to Plus and Team tiers to follow. It represents a step toward practical autonomous research agents built on OpenAI's reasoning model infrastructure.
Import AI 455: AI systems are about to start building themselves
Import AI issue 455 covers the emerging trend of AI systems automating AI research, framing it as a first step toward recursive self-improvement. The commentary synthesizes recent developments suggesting AI is beginning to participate meaningfully in its own development pipeline. As a tier-2 newsletter, this represents curated analysis of frontier AI research directions rather than primary reporting.
Multi-turn evaluation reveals deep research agents fail to compound gains from process-level feedback
A new arXiv paper evaluates deep research agents (DRAs) across multiple feedback turns, comparing self-reflection against process-level feedback via a novel method called Research Gap Inference (RGI). Key findings: self-reflection yields negligible net improvement, one round of process-level feedback raises normalized scores by 8-15 points (~35-40% incorporation rate), but gains do not compound across turns as agents regress on up to 24% of previously satisfied criteria. The results suggest reliable multi-turn improvement remains out of reach for current DRA architectures, highlighting a fundamental limitation in iterative agentic research workflows.
The Age of Async Agents — Cognition's Walden Yan & OpenInspect's Cole Murray
A Latent Space podcast episode featuring Cognition's Walden Yan and OpenInspect's Cole Murray discussing the current state of autonomous software engineering agents. Topics include Devin's reported 80% commit rate, spec-to-PR workflows, full VM environments for agents, agent memory, and the emerging pattern of product managers shipping code directly. The conversation covers practical deployment patterns and tooling for async agentic coding workflows.