From the workshop.
Build logs, architecture notes, and research from the team building SprintEngine Studio.
We Mined 99 of Our Own Sprint Runs. Here's Where Agents Actually Trip Up.
Sprint Engine v1 ran itself: 99 sprints, 889 tasks, and 1,986 review-gate verdicts, each recording structured error counts, quality scores, and friction reports. We archived the v1 run store and analyzed all of it. The findings reshaped what Sprint Engine v2 is.
- Architecture
Fable 5 Plans and Reviews. Opus 4.8 Builds.
Fable 5 returns as the most capable model, at 2× Opus 4.8's price. Put it where judgment compounds — the architect and the reviewer gates — and let Opus 4.8 do the high-volume implementation. You keep Fable-grade planning and review for about a third less than running Fable on everything.
7 min read - Build Log
CI Runs Your Tests. Automations Runs Your Agents.
How Automations runs your agents for you: scheduled and event-driven runs, isolated per-run worktrees, a review-only safeguard, and triggers and actions wired into GitHub, Jira, Switchboard, Watchtower, and Sprint Engine.
7 min read - Build Log
Why Every Sprint Engine Studio Theme Is 6-Bit: Designing for PWM and Dithering Sensitivity
An accessibility decision behind SE Studio's themes: temporal dithering, PWM-sensitive communities, and why authoring every theme color on a 6-bit grid stops our UI from triggering frame-to-frame flicker.
6 min read - Build Log
Inside Sprint Engine: Automating a Sprint Without Losing the Plot
How Sprint Engine automates a sprint end to end: a strategist brief, an architect task graph, specialist implementation, and parallel review gates that produce evidence instead of a transcript.
7 min read - Architecture
Everything Is a Plugin: The Sprint Engine Studio Module System
A look at how SE Studio is built around typed plugin slots — workspace types, roles, providers, skills, and MCPs — so the environment bends to your workflow instead of the other way around.
6 min read - Architecture
Bring Your Own CLI: One Command Contract for Every Agent
Why SE Studio is runtime-agnostic: built-in presets for Claude Code, Codex, and OpenCode, plus a tiny command contract so you can bring your own CLI and keep the subscriptions you already pay for.
6 min read - Build Log
Workspaces and Specialists: Why We Spawn a Team Instead of One Agent
How SE Studio workspaces and specialist roles work, and the data from our own swarm runs that convinced us to spawn a scoped team instead of pointing one generalist agent at everything.
6 min read - Research
The Knowledge Graph: Slim Context for Sharp Agents
How the SE Studio Knowledge Graph keeps agent context slim: small linked notes, an indexed graph, and a worked brand-folder example that shows agents loading only the neighborhood they need.
7 min read - Build Log
Debug and Diagnostics: Seeing Why an Agent Run Went Sideways
A look at SE Studio's debug and performance diagnostics modes: replaying an agent run from its recorded evidence, and a waterfall view that surfaces the slow, expensive spans.
5 min read - Architecture
Bring Your Own Agent Runtime
Why SE Studio is moving toward a runtime-agnostic agent model: built-in Codex and Claude Code presets, custom local runtimes, and a small start/resume contract.
5 min read - Build Log
Switchboard and Watchtower: Local Agent Orchestration You Can Actually Supervise
A build note on how Switchboard and Watchtower move SE Studio from agent sessions to a durable workflow for intake, implementation, testing, review, and evidence.
8 min read - Research
Memory Systems Are the Missing Layer for Agentic Code
A field note on using graph-shaped markdown memory to make AI coding agents faster, more precise, and more useful across very large systems.
8 min read - Architecture
Sprint Engine Studio and Symphony: Building the Command Center for Agentic Delivery
We are making SE Studio Symphony-compatible so teams can supervise agentic software delivery with richer visibility, review, and proof.
7 min read