<- all tokdocs
2026.08.25@alanonai

Vercel Functions Hit 1,800 Seconds Two Months Before This Clip Called Long-Running Tasks a Vercel Tradeoff

Coding
Watch on TikTok

View on TikTok ->

This is an 87-second clip (1:27), shot vertically at 1080x1920, posted 2026-08-25 at 13:32:30 UTC by @alanonai, whose TikTok channel nickname is "Alan on AI". At capture on 2026-10-05 it showed 61,700 views, 3,096 likes, 276 comments, 104 reposts and 1,366 saves. The licensed audio bed is "Embers on the Ivory" by Louise Baker, from the album "Old Letters in a Drawer". I read 23 of the 44 extracted frames (2-second intervals) and all 299 transcript words. There is no screen recording, dashboard, terminal, config file or code anywhere in the 87 seconds. Every frame is the same static selfie shot: a grey-haired man in a dark green henley shirt on a covered porch, with a brown five-blade ceiling fan and frosted globe light above him and a white-framed window behind his left shoulder. The only non-caption graphic is the opening title card on frame 1, which reads "Engineering for AI Vibe Coders" in two lines at the top and "Vercel" in oversized type across the middle. Everything else on screen is burned-in auto-captions, and they carry visible transcription errors: frame 3 reads "just like there's a good chance that your database in the office" where the spoken line is "your database and auth is handled by Supabase"; frame 4 corrects to "handled by Supabase."; frame 7 reads "more than any others, mainly because they're simple to set up and simple to use."; frame 9 reads "So Vercel is one of those."; frame 10 reads "And it's a great way to provide an execution environment to a prototype"; frames 13 and 14 read "AI likes to provide options that are simple to use,"; frame 16 reads "especially for people that don't understand the technology."; frames 19 and 20 read "is the one that is providing the web application to your customers."; frames 25 and 26 read "and planning to use long running a background task."; frame 28 reads "I put a article about Vercel in my engineering for vibe coders series." with "a article" for "an article"; frame 29 reads "and it includes a table of things to consider when you're using Vercel."; frame 31 reads "Now, one thing to remember is that just because your app is running in Vercel"; frame 34 reads "And it doesn't mean that it needs to be tied to vercel" with a lowercase "vercel"; frame 36 reads "so you can work with your AI to add Sections that are not running in Vercel." with a stray capital S; frame 38 reads "And also to make an abstraction layer"; frame 41 reads "when you're app begins to grow." with "you're" for "your"; frame 44 closes on "Let me know what you think and let me know if it's helpful." The hashtags live in the post description rather than a structured field: #ai #vibecoding #coding #programming.

The clip names exactly one Vercel tradeoff, and that tradeoff was largely closed before it posted

The spoken argument has one technical claim in it. At 00:36 to 00:50 the narrator says Vercel is "a great environment to start with, but there are some trade-offs that you need to know about, especially if you're planning to scale and planning to use long-running background tasks." The post description extends the same idea: "A long-running AI process? A background job? A persistent connection? Now we need to think about it."

Correction on chronology. Vercel shipped first-party answers to all three of those workloads before 2026-08-25, and the clip names none of them.

Vercel Queues entered public beta on 2026-02-27, six months before the post, with durable topics, at-least-once delivery and automatic retries, billed per API operation starting at $0.60 per 1M operations. The Queues docs list the availability as "Vercel Queues (Beta) are available on all plans" and name "Defer expensive work" and "Absorb traffic spikes" as the intended uses.

Function duration moved on 2026-06-15, ten weeks before the post. Vercel's changelog records that Pro and Enterprise functions went from an 800-second ceiling to 1,800 seconds (30 minutes) on Node.js and Python, with durations above 800 seconds in beta and requiring maxDuration: 1800 in the route config or vercel.json. The current Vercel Functions limits page confirms the table: Hobby gets "300s default and maximum", Pro and Enterprise get "300s default, 800s maximum, and 1800s extended maximum Beta".

Persistent connections landed on 2026-06-22, two months before the post. WebSocket support went to public beta, with Vercel Functions serving WebSocket connections directly, Socket.IO supported, and connections following the same limits and pricing as other function invocations on Fluid compute.

For work that exceeds any duration cap, the limits page points at Vercel Workflows, which it describes as the path "for workloads that require unlimited execution time", letting code "pause, resume, and maintain state for minutes to months without duration limits".

None of this makes the clip's advice wrong. It makes the stated reason for the advice stale by two to six months at the moment of publication.

The 10-second serverless timeout that vibe coders repeat applies only to pre-April-2025 projects

The folklore version of this topic is a 10-second Hobby function limit. Verified against the source, that number exists but is legacy.

The Vercel limits page states the condition explicitly: "If you have an existing project, deployed to Vercel before April 23rd 2025 and not using Fluid compute", the defaults and maximums are Hobby 10s default / 60s maximum, Pro 15s / 300s, Enterprise 15s / 900s.

Any project created after that date runs on Fluid compute by default, and the table flips to 300s default on every plan. A vibe-coded app deployed in 2026 starts with a five-minute function ceiling on the free tier, which is thirty times the legacy Hobby default.

Verified limits that do apply to every plan equally, from the same page: build time per deployment is capped at 45 minutes on Hobby, Pro and Enterprise alike, the proxied request timeout is 120 seconds on all three, and cron jobs cap at 100 per project on all three. Hobby is held to 100 deployments created per day and 1 concurrent deployment, against 6,000 per day and up to 500 concurrent builds on Pro.

Hobby's binding constraint is the non-commercial rule and a 30-day lockout, which the clip's own framing walks into

At 00:30 the narrator says that once your app is deployed, "Vercel is the one that is providing the web application to your customers." Customers are the problem on the free tier.

The Vercel Hobby plan doc states it plainly: "As stated in the fair use guidelines, the Hobby plan restricts users to non-commercial, personal use only." An app with paying customers does not belong on Hobby under Vercel's own terms, regardless of whether it fits inside the quotas.

The failure mode when you exceed quota is also harsher than most people expect, and it is not a bill. The same page: "As the Hobby plan is a free tier there are no billing cycles. In most cases, if you exceed your usage limits on the Hobby plan, you will have to wait until 30 days have passed before you can use the feature again." Web Analytics is called out as a shorter exception, resuming after 7 days.

Verified Hobby allotments from that page: 100 GB Fast Data Transfer, 10 GB Fast Origin Transfer, 1,000,000 CDN Requests, 1,000,000 Function Invocations, 4 CPU-hours of Active CPU, 360 GB-hours of Provisioned Memory, 5,000 Image Transformations, 200 projects, 2 build vCPUs and 8 GB build memory. The pricing page lists Hobby at $0/mo and Pro at $20/mo, with Pro moving Fast Data Transfer to 1 TB per month and CDN Requests to 10M per month. The Hobby doc's upgrade instructions price Pro seats at "$20 per user / month, while Viewer seats are free".

Spend controls are real, Pro-only, and off by default in the sense that matters

The clip makes no cost claim, so there is nothing to correct on that front. The surrounding genre does, and the documented behavior is worth stating precisely rather than leaving to folklore.

Vercel's Spend Management doc carries an availability banner reading "Spend Management is available on Enterprise and Pro plans (Enterprise teams require the Flexible Commitment plan)". The Hobby comparison table on the Hobby plan page lists Spend Management as "N/A" for Hobby. A Hobby project cannot run up a bill, because it gets paused instead.

The caveat that actually bites sits in the doc as a highlighted note: "Setting a spend amount does not stop usage on its own. To stop incurring costs above your spend amount, turn on Pause Production Deployments so Vercel pauses your projects when the amount is reached." A second note adds that the check "happens every few minutes", so "projects can keep serving traffic and accruing usage for several minutes after you cross the spend amount."

Two Vercel changelog entries describe the default differently. Spend Management now pauses production deployments by default is dated 2024-06-27, while Spend Management now enabled by default on Pro, dated 2025-09-09, says alerts fire but "your deployments will continue without interruption unless a hard limit is manually configured." I have not reconciled those two, and I am not going to assert which default is live. The doc page, last updated 2026-09-18, is the version to trust, and it says to verify the pause toggle yourself.

On the underlying meter: Active CPU pricing for Fluid compute shipped 2025-06-25 at $0.128 per Active CPU hour and $0.0106 per GB-hour of provisioned memory. The functions limits page explains the consequence that matters for AI apps: "Waiting for I/O (e.g. calling AI models, database queries) does not count towards active CPU time." A function that sits blocked on an LLM call for four minutes bills CPU for a fraction of that window.

The article the clip sends you to is not one I could find

At 00:51 the narrator says "I put an article about Vercel in my Engineering for Vibe Coder series. You'll find a link to it in the comments, and it includes a table of things to consider when you're using Vercel."

Verified: @alanonai is Alan Knox. alanknox.com links the @alanonai handle and uses the "Alan on AI" branding, and the site hosts the "Engineering for Vibe Coders" series.

Unverified: I could not find a Vercel-specific entry in that series. Searches surfaced series articles on messaging queues, long-running processes, rate limiting, graceful degradation, notifications, data contracts, business requirements, development environments and deployment targets. A direct request to alanknox.com/vercel-engineering-for-vibe-coders/ returned HTTP 404. The nearest match, Deployment Targets: Engineering for Vibe Coders, dated 2025-11-13, discusses VMs against serverless in general and does not name Vercel or quote any platform's limits. The article may exist at a URL I did not find, or it may be newer than what search indexed. I am reporting what I could and could not confirm rather than assuming it is there.

Two smaller naming corrections while I am here. The on-screen title card says "Engineering for AI Vibe Coders", the post description says "Engineering for Vibe Coders: Vercel", the spoken line says "Engineering for Vibe Coder series", and Alan Knox's own site calls it "Engineering for Vibe Coders". The last one is the author's canonical spelling. Separately, the claim at 00:05 that "your database and auth is handled by Supabase" for most vibe-coded apps is not sourced in the clip and I found no primary measurement of it, so treat it as the narrator's impression rather than a statistic.

Key Takeaways

  • Correction: The clip's single named tradeoff, long-running background tasks, was addressed by Vercel before the 2026-08-25 post date. Queues hit public beta 2026-02-27, 1,800-second functions shipped 2026-06-15 for Pro and Enterprise, and WebSocket support hit public beta 2026-06-22.
  • Correction: The 10-second Hobby function timeout applies only to projects deployed before 2025-04-23 that are not on Fluid compute. New projects get 300s as both default and maximum on Hobby, and 300s default with an 800s maximum on Pro and Enterprise.
  • Correction: The on-screen title card reads "Engineering for AI Vibe Coders", which does not match the series name on the author's own site, "Engineering for Vibe Coders".
  • Partial correction: The advice to move work off Vercel is sound for some workloads, but the clip skips Vercel Workflows, which the functions limits page names as the supported path for workloads requiring unlimited execution time.
  • Verified: Hobby is $0/mo with 100 GB Fast Data Transfer, 1M CDN requests, 1M function invocations, 4 Active CPU-hours and 360 GB-hours of provisioned memory per month. Pro is $20/mo with 1 TB transfer and 10M CDN requests, and developer seats cost $20 per user per month.
  • Verified: Build time per deployment caps at 45 minutes on every plan, including Enterprise.
  • Unstated cost: Vercel's Hobby plan "restricts users to non-commercial, personal use only." The clip's own framing of Vercel "providing the web application to your customers" describes a use that Hobby's fair-use guidelines do not permit.
  • Unstated cost: Exceeding a Hobby quota does not produce a bill. It locks the feature for 30 days in most cases, with Web Analytics resuming after 7 days.
  • Unstated caveat: Spend Management is Pro and Enterprise-Flex only, and setting a spend amount does not stop usage by itself. Pause Production Deployments has to be turned on, and the spend check runs every few minutes, so traffic keeps accruing after the threshold is crossed.
  • Unverified: I could not locate a Vercel-specific article in the Engineering for Vibe Coders series. The direct URL returns 404 and the closest published entry, Deployment Targets (2025-11-13), does not mention Vercel.
  • Unverified: The claim that most vibe-coded apps use Supabase for database and auth is not sourced in the clip and I found no primary data for it.
  • Caption quality: The burned-in auto-captions contain at least five errors, including "your database in the office" for "your database and auth is handled by Supabase", "I put a article", "tied to vercel", "long running a background task" and "when you're app begins to grow."

Resources

Published August 25, 2026. Writeup generated from a favorited TikTok.