Software is made between commits

Reception of the DeltaDB Concept

• Some see DeltaDB’s “every operation between commits” as a natural substrate for AI agents and collaboration, especially for remote, agent-heavy teams that already treat LLMs as primary coding partners.
• Others view it as intrusive “keylogging for code,” adding noise, cognitive load, and review burden with little clear benefit over well-crafted commits and PRs.
• Several people suspect it mainly benefits AI training and enterprise surveillance/audit needs rather than day‑to‑day human developers.

Commits vs. Fine-Grained History

• One camp values frequent, granular commits and rich history for git blame/bisect, debugging, and understanding evolution.
• Another prefers aggressively cleaned-up, rebased or squashed histories that tell the “theoretical story” of the code, not the messy path taken.
• Many argue the useful unit is a small, atomic commit or change list, not keystroke-level deltas; they liken DeltaDB to recording every scratch note or draft, which is rarely revisited.
• Some note existing tools (Git with frequent commits, --first-parent, Gerrit/Phabricator, jj) already support fine-grained history or autosave-like snapshots without centralizing everything.

AI Agents and Coding Workflow

• A subset of commenters already use LLMs extensively, with agents writing most of the code and humans guiding and reviewing. For them, integrating agent conversations and code evolution in one place is appealing.
• Others argue the right answer is better documentation, comments, and PR descriptions, not replaying agent chats and intermediate deltas.
• There’s concern that tooling is being redesigned around agents rather than human collaboration needs.

Privacy, Surveillance, and Workplace Control

• Strong unease about employers owning a complete record of “thinking between commits,” including rants, confusion, and temporary secrets.
• Some predict mandatory AI provenance logs, token/impact metrics, and audit trails used for performance reviews, layoffs, and legal defense.

Zed as Editor vs. Platform

• Many praise Zed as fast, minimal, and distraction-free, valuing its current ability to disable AI features.
• Others worry about a shift toward “Cursor-like” agent-centric workflows, bloat, and dependence on Anthropic/OpenAI; forks like Gram are welcomed as “AI-free” insurance.

Collaboration and PRs

• Some agree PRs are a late, coarse collaboration point but see DeltaDB as the wrong fix; they’d prefer better support for small changes, stacked/atomic review, and multi-repo workflows.
• Overall, there is no consensus that per-operation versioning materially improves human collaboration over improving existing VCS and review practices.