Key findings
- External agents are the most raised concern. Of 86 accounts, 17 say Claude Code, Codex, or another ACP agent lacks commands, history, or context controls inside Zed, and Zed's own documentation states that panel features vary by agent.
- The Zed Agent is the most praised feature. Of the accounts that judge it, 16 are positive and one is not, citing its review flow, parallel threads, and model switching mid-task.
- Completion splits buyers. Six accounts find Zed's edit predictions behind Cursor Tab and two praise them, so completion needs its own pilot.
GILD read public accounts of Zed's AI features published May 7, 2025 to October 16, 2026, along with Zed's documentation. The aim is to answer what a buyer wants to know before moving AI coding into Zed.
What the record covers
The record holds 55 Hacker News comments, 7 issues filed on zed-industries/zed, 6 blog posts, 17 posts on X, and a podcast transcript. Most Hacker News accounts come from threads under Zed's own AI announcements: the agent panel launch, Claude Code and Codex over ACP, token pricing, parallel agents, and Zed 1.0.
Eighteen accounts speak for a team or a technical leader. Most come from named CTOs and founders, and the others include a corporate buyer and a technical director. None describes an organization-wide rollout. Reddit refused access to the threads GILD found, so none is in the record.
Agents: Zed's own or yours
Accounts that pick an agent mostly bring their own. Of the 27 that take a side, 21 run Claude Code, Codex, or another external agent in Zed's panel or a terminal. The other 6 prefer the Zed Agent with a chosen model. Several external-agent users still like running the agent beside the code.
Zed's documentation says history, checkpoints, and token usage display depend on each external agent's integration. A team should list the commands its engineers use and check each one in the panel.
Cost and model access
Zed bills hosted usage at API list price plus 10% after Pro's US$5. Accounts that discuss paying split between Zed's plan, for edit predictions and one bill, and their own keys, gateways, or company tenants to avoid the markup or keep governance in place.
Nine accounts praise model choice, including local Ollama models and OpenRouter.
What to test
Run the same tasks through the Zed Agent, each external agent in the panel, and the agent's own app. Pilot edit predictions on your own repositories against the tool engineers use today, and watch for the crashes, freezes, and slow agent runs that 9 accounts report.
Methodology
GILD searched Hacker News through the Algolia API, issues on zed-industries/zed, the web, and X. It read each kept page on October 25, 2026 and archived an exact copy. Each record is coded only from an exact quote in its archived page, with author and publication date taken from the page or API. GitHub issues count as mixed. The record leans toward people who read Zed's announcement threads and is not a representative survey.
Products in this piece
Sources
- Official documentation
- 8
- Public source
- 6
- Zed docs: Agent Panel
- Zed docs: External Agents
- Zed docs: Edit Prediction
- Zed pricing
- Zed docs: LLM Providers
- Zed blog: Zed is 1.0 (Apr 29, 2026)
- Zed blog: Claude Code via ACP
- Zed blog: Introducing Parallel Agents in Zed (Apr 22, 2026)
- Alek Dobrohotov: Zed + Claude Code, a beautiful promise not quite there yet (Sep 6, 2025)
- Victor Skvortsov: I switched from VSCode to Zed (Jan 3, 2026)
- Braw.dev: Reallocating a US$100 per month Claude Code budget to Zed and OpenRouter (Apr 6, 2026)
- Hacker News, sodacanner: what Claude Code lacks inside Zed (Apr 29, 2026)
- Hacker News, zephod: back to Cursor for Cursor Tab (Apr 30, 2026)
- GitHub issue #55501: Claude Code ACP session stuck after a usage limit (May 2, 2026)