--- title: 'In the News: August 13, 2026, Evening' description: 'DeepSeek open-sourced its coding agent harness under MIT, with a forkable, replayable session log and a benchmark mode built to strip harness effects out.' canonical_url: 'https://darkfactory.dev/news/2026-08-13-evening' markdown_url: 'https://darkfactory.dev/news/2026-08-13-evening.md' collection: news date_published: '2026-08-13T20:05:00-04:00' date_modified: '2026-08-13T20:05:00-04:00' --- # In the News: August 13, 2026, Evening DeepSeek shipped its coding agent harness under MIT today with an append-only log of everything the model saw, which users can fork and replay, plus a stripped-down mode for measuring a model without harness assistance. Zed separately announced an application built on the premise that the record of how agent-written code came to be should live between commits rather than in them. ## 1. DeepSeek open-sourced its coding agent harness, and made the trajectory a first-class object **[DeepSeek Harness developer preview: Everything is a plugin](https://deepseek.com/harness/en/)** · DeepSeek AI · Announced August 13, 2026 DeepSeek released `dsh`, its own coding agent harness, as a developer preview with the source under an MIT license. The design claim is that every capability is a swappable plugin: "models, tools, skills, sessions, sandboxes, storage, loops, scheduling, and the UI", all mounted on a kernel called Cordis, and all replaceable from configuration without touching the harness source. The landing page opens with the equation `AGENT = MODEL + HARNESS` and glosses it: "The model is the soul of an agent. A harness lets an agent understand its environment, use tools, and keep working in real-world settings." That is a frontier lab defining the primitive in the first person. The instrumentation is explicit: "Everything the model sees is recorded in an append-only session log: system prompts, reasoning, tool calls and results, subagent scheduling, and every context injection." A Trajectory view reads those records by source, and "resume, fork, search, and replay all operate on the same event stream." Minimal mode provides a runtime built for measurement rather than work: a two-tool agent with persistent bash and `str_replace_editor` only, kept, in DeepSeek's words, "for benchmarking models in a minimal environment". The company's own API changelog shows it using that mode as the framework for the official V4-Flash benchmark numbers published on July 31, with the effort level, top-p and temperature all stated. The preview warns in capitals that there will be compatibility-breaking changes. The repository stood at 38,600 stars and 3,000 forks when checked at about 7:45pm Eastern on August 13, hours after release, and the submission sat at the top of Hacker News for most of the day. The Cordis paper the README cites was not read for this item. **Why it matters:** Treating the session log as the single event stream that resume, fork and replay all read from is a concrete answer to a problem most harnesses handle with ad hoc transcripts. It separates reconstructing a run from re-running it at a chosen point. Minimal mode makes a second point: by shipping a mode that removes its own harness from a measurement, the vendor concedes that harness and model results are not separable by default. Most published agent benchmarks assume otherwise. ## 2. Zed built a new application around the idea that review belongs between commits **[Introducing Delta](https://zed.dev/blog/introducing-delta)** · Nathan Sobo, Zed Industries · Zed Blog, August 12, 2026 Delta is a separate application from Zed, in private beta as of yesterday, built on DeltaDB, the version control layer Zed has been developing publicly. The argument is about where the record of agent work lives. "On a commit-based platform, comments attach to snapshots and fall out of date as soon as the code changes." DeltaDB instead captures every edit and conversation between commits, replicating the worktree and the thread together, while leaving the underlying git repository intact: "teammates who never open Delta see a normal git repo." The practical consequence Zed is aiming at is review. Comments anchor to lines in the evolving worktree rather than to a diff snapshot, the agent stays in the thread with the conversation that produced the code, and, in Sobo's framing, "you don't reconstruct intent from a diff. You ask the agent to explain it or fix it." The post is also blunt about interface load: "Agents produce more text and bigger changes than any human, and most tools cope by hiding things (e.g. collapsing diffs, truncating transcripts, summarizing what you should be able to read)." Delta renders diffs and transcripts in full and treats the conversation as a document you can put a cursor anywhere in. It connects to third-party agent harnesses, starting with Claude Code, so a terminal session syncs into a shared thread. This is an announcement with a waitlist behind it, not something you can evaluate today. Invites went out on the 12th with more promised over the coming weeks. **Why it matters:** If the argument holds, the reviewable artifact for agent-written code stops being the diff and becomes the conversation plus the operation stream that produced it. That is a claim about what a team should be keeping, and it is testable against your own practice now: for the last non-trivial change an agent made in your codebase, ask whether the commit contains enough to review it, or whether the reasoning you would need is in a chat window nobody archived. --- ## Also this cycle - **[Harness engineering for coding agent users](https://martinfowler.com/articles/harness-engineering.html)** · Birgitta Böckeler, Thoughtworks, on martinfowler.com · Not new, published April 2, 2026, and still the clearest map of the vocabulary the day's release is speaking. It splits an agent harness into guides that steer before the agent acts and sensors that observe after, each either computational (linters, type checkers, structural tests) or inferential (review agents, LLM as judge), and argues the human's job is to iterate the harness rather than the code. The article notes in its own text that it supersedes Böckeler's February memo and that the memo's URL now redirects here, which matters if you have the older piece bookmarked or cited. --- ## Thread watch _Discussions gathering force. No primary read yet, so these are reported as discussions, not as findings._ - **[AI agents lie, cheat and steal. That is putting off users](https://news.ycombinator.com/item?id=49285604)** · Hacker News, on an Economist article · The Economist piece is paywalled and was not read for this edition, so nothing is reported here about what it argues. The discussion around it is the notable part: roughly 150 points and 188 comments by early evening Eastern on August 13, with an unusually high comment-to-point ratio, and the argument is definitional. Commenters are split on whether a harness primarily restrains a model or primarily extends it, with one formulation, that the harness supplies both the fence and the bolt cutters, capturing why the question stays open. Worth watching because the same dispute is what the day's release answers by fiat.