Founders test Meta's Muse Spark: fast but needs heavy hand-holding

After Meta released the Muse Spark coding model alongside Muse CLI, leaders of developer-tool companies including Sourcegraph, Zed and Sentry shared early hands-on impressions on X. The consensus: it is fast with decent output quality, but weak at following preset instructions and apparently light on reasoning tokens, requiring heavy manual guidance. Bottom line: Muse Spark is an interesting early product, but its agentic capabilities remain a clear weak spot.

Confirmed

Sourcegraph CEO Quinn Slack found both the model and Muse CLI pleasant to use, and praised Meta; he added that Muse Spark is gentler than models like Fable in following native steer without being overbearing, which he sees as a virtue given the solid output quality.
Zed founder Nathan Soller reported in testing that the agent largely ignores preset instructions such as AGENTS.md, and noted it spends few reasoning/thinking tokens, which he suspects is related.
Sentry CEO David Cramer gave a similar verdict: fast but requiring heavy hand-holding, likewise citing weak AGENTS.md adherence and low reasoning-token usage.
Soller also complained about an aggressive sandbox policy: with no network access inside the sandbox, the agent could not authenticate CLI tools automatically, forcing a manual workaround he described as a love-hate annoyance.

Why it matters

These are early, first-hand reactions from leaders of frontline tool vendors rather than marketing reviews, making them highly credible baselines.
The feedback is consistently split: "fast" and "pleasant" coexist with "needs hand-holding," indicating clear gaps in instruction-following and autonomous reasoning that will be worth tracking in future iterations.

2026-09-12 ~ 2026-09-12 · 5 related posts

Primary sources

Sourcegraph CEO Praises Meta's Muse Spark Coding Model and CLI — zeeg · 2026-09-12 07:58 (UTC+08:00)
Zed CEO on Meta's coding agent: fast and pleasant, but needs a lot of hand-holding — zeeg · 2026-09-12 08:56 (UTC+08:00)
Sentry CEO on Meta's Muse Spark: fast and pleasant but skimps on reasoning, needs heavy hand-holding — zeeg · 2026-09-12 08:30 (UTC+08:00)

[source] Sourcegraph CEO Praises Meta's Muse Spark Coding Model and CLI — zeeg · 2026-09-12
Muse Spark Less Aggressive at Following Prompts Than Fable, but Quality Holds — zeeg · 2026-09-12
[source] Sentry CEO on Meta's Muse Spark: fast and pleasant but skimps on reasoning, needs heavy hand-holding — zeeg · 2026-09-12
[source] Zed CEO on Meta's coding agent: fast and pleasant, but needs a lot of hand-holding — zeeg · 2026-09-12
Zed founder on agent sandboxes: aggressive isolation means manual auth steps — zeeg · 2026-09-12