"Cancelling Cursor": Quality debt takes precedence over feature velocity

Ongoing story : Fatigue hype 2026 : le tri entre modèle et harness· Part 15/15

CraftSubscribers only 38 min ago0Add to bookmarks

"Cancelling Cursor": Quality debt takes precedence over feature velocity — Illustration : Léa Fontaine

Alex Yumashev (Jitbit) explains why he unsubscribed from Cursor: VSCode base 7-8 months behind, 7 bugs reported in a year and never fixed, Agent mode forced with every update, ghost support. An individual signal that perfectly illustrates the editor-orchestrator divide.

In plain terms. Jitbit founder Alex Yumashev has published a detailed post outlining why he canceled his Cursor subscription: outdated VSCode base, unresolved legacy bugs, default shift to agent mode, and failing support. Cursor pioneered inline diffs and contextual chat, now industry standards, but its pivot toward agent mode and growing divergence from upstream VSCode have fueled months of community backlash.

Context

Cursor popularized inline diffs and contextual chat, now industry standards. Its product pivot toward agent mode and increasing divergence from upstream VSCode have fueled months of community backlash.

The facts (from the post)

Alex Yumashev, Jitbit founder and long-time paying customer:

VSCode base: ~7–8 months behind upstream. Some recent extensions fail to install.
Bugs: ~7 issues reported over a year, none fixed. Example: a CSS defect in settings reported 4 months ago.
Agent by default: each update restarts the app in Agent mode. The toggle to disable it is moved and renamed every release.
Support: a critical Reddit post (177 upvotes) was removed by moderators. Tickets ignored or routed to AI-generated responses.
Alternative cited: Claude Code, presented as usable directly inside Cursor.

Analysis

A single signal, a broader trend. Cursor’s default shift to agent mode mirrors the divide Rachel Laycock identified (#1752): dev-orchestrator vs. dev-editor. Cursor chose orchestrator; part of its base wanted an editor.

Craft angle: when a vendor leaves 7 bugs open for a year in a subscription product, the question shifts from speed of innovation to quality debt. Yumashev frames it as support he’d expect from a side-project open-source tool—the implicit craft-SaaS contract is broken.

Under the hood - platform debt

Cursor forks VSCode and delays upstream sync. At 7–8 months behind, the gap blocks recent extensions. This isn’t a one-off bug; it’s platform debt accumulating with every VSCode release.

Probabilized scenarios

Cursor pivots “back to IDE” (~30%): reversibility possible under community pressure.
Cursor doubles down on agent mode, loses craft base (~50%): segmentation into a product distinct from historical IDEs.
Fragmentation toward Claude Code / Codex / native VSCode (~20%): the “IDE + pluggable agents” stack replaces the closed product.

Implications for pros

If you buy a dev tool “agentic-first,” ask for the SLA on cosmetic bugs open 3+ months. That’s the strongest predictor of perceived quality over time—and what separates a craft tool from an orchestrator tool.

Watchlist

Cursor MRR and churn; official Cursor response to the post; standalone Claude Code adoption; positioning from other legacy editors.

So what. The agent-first pivot fractures the dev-tools base. What matters when choosing: the rate of old bugs left open, not the count of new features.