AutoResearch as a Product-Building Loop
AutoResearch is not one feature. It is the running attempt to make EdgeLab capable of evaluating, changing, and improving parts of itself through agent-driven loops.
AutoResearch as a Product-Building Loop
AutoResearch is not one feature. It is the running attempt to make EdgeLab capable of evaluating, changing, and improving parts of itself through agent-driven loops.
Series: autoresearch March 29, 2026
What this is
EdgeLab from Karthik's lens, with the work left intact.
This publication is a living stream of EdgeLab development: the larger research arcs, the smaller engineering notes, the systems that stabilize over time, and the factual record of using agents to help build the product.
Publication model
- `Journal` holds the reverse-chronological stream of posts, notes, and updates.
- `Series` groups long-running work like autoresearch into coherent arcs.
- `Systems` stays separate as the durable reference layer.
- Each piece is factual first, with style and argument refined later.
Latest journal entries
Async Job Processing Is Still the Core Boundary
Most of the platform complexity starts from one durable requirement: request-response paths must stay thin while real work moves through queues and long-running workers.
Using Agents to Build EdgeLab Is Now Part of the Record
The publication needs to document not only the product architecture but also the fact that agents like Codex and Claude are participating in the development workflow.
Active series
Ongoing arcs instead of disconnected posts.
AutoResearch
The growing body of work around letting agent loops evaluate, mutate, and improve strategy-building workflows inside EdgeLab.
2 entries
Engine Foundations
Notes and deep dives around the execution model, job processing, and service boundaries that keep the platform moving.
1 entry
System reference map
Durable references for the parts of EdgeLab that have stabilized enough to document.
Agent-Assisted Build Loop
A systems reference for how agent help enters the EdgeLab workflow: context gathering, bounded implementation, verification, and human review.
Workflow and Job Execution
A reference page for the async execution model that separates user-facing request paths from the heavier work done in background services.