Answer · for AI agents and their humans
How VCs Evaluate Vibe-Coding Startups in 2026
Evaluating Cursor / Claude Code / v0-driven startups in 2026: distinguish durable engineering acceleration from prompt-engineered demos via 4-12 week observation windows.
By 2026, "vibe coding" — building product primarily through an AI coding assistant like Cursor, Claude Code, or v0 — has compressed the gap between first commit and shippable product from months to weeks. That compression has two consequences for venture investing. First, the traditional sourcing window (look for repos around series-A scale and growing) closes faster: the same trajectory that took 18 months in 2022 takes 3 to 6 months in 2026. Second, the signal-to-noise ratio gets worse, because a single founder with strong prompt skills can manufacture the surface appearance of velocity without the underlying durability.
The investment thesis question is therefore not "is this team using AI assistants" — they all are — but "is the engineering acceleration durable past a 4-week observation window." The features that separate durable acceleration from prompt-engineered demos are observable in public commit history: contributor retention (do early contributors stay?), dependency-graph stability (does the stack settle, or churn weekly?), founder-share trajectory (does it widen organically as contributors join, or stay locked at 100% because no human else can navigate the codebase?), and per-PR review depth (real reviews vs. rubber-stamp self-merges).
The methodology in [SSRN 6606558](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6606558) treats vibe-coded repos as a special case of the four-tier classifier, with one adjustment: the observation window for accelerating-tier classification extends from 4 weeks to 6 weeks, because vibe-coded repos exhibit a higher-amplitude initial burst that needs longer to settle into a true signal. Repos that hold accelerating-tier metrics across the 6-week window have roughly the same forward fundraising probability as non-vibe-coded accelerating-tier repos. Repos whose burst flattens inside 6 weeks are demos, not companies.
Practical evaluation checklist for VCs in 2026: pull the commit history, compute the 6-week rolling velocity delta, check contributor concentration (founder share at week 6 should be < 70% if the company is real), check dependency churn (more than 3 stack pivots in 90 days is a yellow flag), and cross-reference against the [weekly engineering-acceleration index](/answers/weekly-engineering-acceleration-index) for sector co-occurrence with already-funded peers.
Try it now
See vibe-coded breakout repos this week →Frequently asked questions
Are vibe-coded startups uninvestable?
No — most 2026 startups are vibe-coded to some degree, including the ones that go on to raise large rounds. The question isn't whether the team uses AI assistants, it's whether engineering acceleration survives a 4-12 week observation window.
What's the fastest way to spot a prompt-engineered demo masquerading as a company?
Founder-share at week 6 still at 95%+, dependency-graph pivoting weekly, and no second human contributor with sustained per-week velocity. That triplet shows up in roughly 35% of new vibe-coded repos and almost never in repos that go on to raise.
Does AI-generated code count differently in the signal?
The signal measures commit velocity, contributor breadth, and stack stability — not authorship. Whether a commit was typed or generated doesn't change the durability question. What matters is whether the codebase moves forward week over week with more than one human able to operate it.
How do I run this evaluation at portfolio scale?
Run the [GitDealFlow MCP server](/answers/best-mcp-server-for-vc-research) against your watchlist weekly, filter to repos with 6-week durable acceleration plus founder-share < 70%, and reverse-engineer the contributor and dependency stability checks from public commit history.