← Back to blog
AiAbout 6 min read

Kino AI Sells Agent Edits by the Call, Not by the Render

Published Nov 14, 2026
Kino AI Sells Agent Edits by the Call, Not by the Render

Kino AI launched an AI media manager and video editor this week, and the product description reads like a list of things a normal editor does: store footage, cut on a timeline, collect review notes. The part that matters is on the side of the timeline. Kino runs a Model Context Protocol server, which means Claude Code, Codex, Cursor and ChatGPT can search the footage and change the project from their own chat windows, without anyone opening Kino's UI.

Kino is a Y Combinator S23 company, and it has been building around a specific frustration. Video production is split across three places that do not talk: footage on a local drive, editing in a heavy desktop app, review in a spreadsheet or a shared link. Kino's answer is to treat the whole project as one shared environment, with humans and agents looking at the same assets and the same timeline.

What the MCP server actually exposes

Two things: the media library and the timeline. An agent can search through footage by description, pull the clip it wants, and place it. It can cut, reorder and set a style. Cloud sharing runs on 720p proxy transcodes so assets stream by URL instead of waiting on a download.

That last detail is the unglamorous engineering that decides whether an agent workflow feels fast. An agent that has to fetch a 4GB camera file before it can look at frame 300 is not an agent, it is a batch job. Proxies make the difference between a two-second check and a two-minute one, and they make it possible for a model to look at forty clips in the time it takes to read this paragraph.

Kino's own framing is careful about the limit. A tool like Claude Code can one-shot a video, the company says, but Kino provides the infrastructure for a human to watch the result, leave notes, adjust a font size and source extra B-roll before delivery. That is a more honest description of where agents are than most launches manage. The generation is not the bottleneck anymore. The review loop is.

The pricing tells you what they think is scarce

The free tier is unusually complete: full editor, unlimited imports and exports, and 100 MCP tool calls per week. Pro runs $20 to $25 a month for unlimited MCP calls, more storage, and the ability to share timeline and media links.

Read that structure backwards. Kino is not metering renders or exports or minutes of footage. It meters agent calls. The company has decided that storage and compute will behave, and that the number of times a model pokes at a project is what will run away from you.

A steel plate machined with a grid of recesses, most holding a small brass counter.

That is a reasonable guess and an instructive one. An agent editing a five-minute cut might make hundreds of small calls: search footage, read the current timeline, check the duration, place a clip, check again. Each call is cheap. A hundred of them per edit, across a team of five, every day, is not. Charging by the call puts the meter exactly where the behaviour is.

It also creates a strange incentive. If calls are the unit, an efficient agent is worth more than a clever one, and a vendor has a reason to keep the tool surface small. Kino lists a bounded set of operations rather than a general "do anything" endpoint. That is the right call for reliability, and it happens to be the right call for the business model.

Why a timeline is a better target than a prompt box

The last two years of AI video tools have mostly attacked generation. Type a sentence, get a clip. The output is impressive and the workflow is a dead end, because the clip arrives as a file, and the file has to be assembled with other files by someone who has opinions about pacing.

An MCP-editable timeline attacks the assembly instead. The pieces can come from anywhere: camera footage, an AI-generated shot, a stock clip, a screen recording. The agent's job is placement and structure, which is the part of editing that is rule-bound enough to automate and boring enough that people want it automated.

This is the same realisation that pushed Adobe to promise MCP integrations for Photoshop, Illustrator, Premiere and After Effects in early 2027, and the one behind YouArt's decision to drive professional editors through their menus. Three different approaches, one shared premise: the valuable surface is the project, not the prompt.

Where Kino differs is that it owns the project format. Adobe and YouArt have to work inside software whose file format someone else controls. Kino can design its data model around what an agent needs to read, which is why it can hand an LLM a structured timeline instead of a screenshot of one.

The risks are the usual ones

A shared environment is only as good as its permissions. If an agent can edit the timeline, it can also delete a take, and "undo" across a collaborative canvas is a harder problem than undo in a single-user app. Kino's review system implies versioning, but the launch material does not describe how far back it goes or who can restore what.

Second, an agent that edits by description will make choices a human would not. It will pick the second-best take because the description matched. It will cut on a beat that is technically correct and emotionally wrong. The review loop is the mitigation, and the review loop needs a human with time, which is exactly what the product is supposed to save.

Third, the MCP call budget creates a ceiling on ambition. Teams that build elaborate agent workflows will hit 100 calls a week in an afternoon and upgrade, which is fine, and then discover that unlimited calls encourage agents to be sloppy. Rate limits work as a design tool as much as a paywall.

What to take from it

Kino is a small company with a sensible thesis: make the video project a programmable object, then let whatever agent is fashionable this quarter operate it. The bet does not depend on any single model winning. It only depends on agents getting good enough to be trusted with a timeline, which is already happening in narrow slices.

The next twelve months will show whether editors accept an agent as an assistant editor who needs supervision, or whether they treat the whole category as a toy that produces unusable first cuts. Kino's free tier is priced to answer that question quickly. A hundred calls a week is enough to find out, and cheap enough that nobody has to argue about it in a budget meeting.

Related articles

Google replaced the model behind Search, and the number that moved was a shell benchmark

Gemini 4 Hydrogen became the default on 12 November, taking over from Gemini 3.8 Flash across Search, the Gemini app and the developer API. It runs about 2.5 times faster, scores 57% on Terminal-Bench 4.1 where its predecessor scored 18.2%, and costs $1.50 in and $7.50 out per million tokens.

Three ways a coding agent wastes money, measured on 30 repositories

TokenEconomizer watches what an agent did in a repository you work in often, then stops it from rediscovering the same build command, loading tools it never uses and looping on one error. Across five agent and model combinations it cut mean cost by 18.4% to 30.9%, cheaper in all 30 repository cells, with Pass@1 up in four of five.

earshot gives coding agents 922 sounds, and every one is rendered from code

An agent can write a game, an interface or a film and never hear any of it. earshot answers requests like "three variations of a footstep on gravel" from a catalogue of 922 designs built from public-domain recordings, with no audio files in the repository at all. It runs on CPU, as a CLI, a Python package, an MCP server and a Claude Code plugin.

The agents stealing API keys were not sophisticated. The gateways were open.

Amazon's decoy servers spent October posing as inference APIs and logged agents built on ARTEX pulling provider keys out of LiteLLM, Dify, Langflow and Open WebUI. The way in was a default master key and a half-finished install. ARTEX itself ships with budgets, scope gates and evidence for every finding.