AI video editing in 2026: the assistant now hands over a project file

AI video editing crossed a line at IBC 2026

AI video editing crossed a line at the RAI in Amsterdam this September. At IBC 2026 the editing vendors did not show better generation. They showed agents that prepare media, and one of them hands the editor a native project file instead of an XML to reimport. That is a smaller change to watch and a larger one to work with, because the handoff between machine and editor is where a schedule is actually won.

The show ran 11 to 14 September, and the post-production news was not about prettier frames. Avid used the week to launch a browser-based Media Composer built on its Content Core platform with Google Cloud's Gemini Enterprise embedded, and to demonstrate Ready to Edit, an agent that takes natural-language prompts to sync media, transcribe and align dialogue, and organise project bins before an editor ever opens the room.

Adobe put a generative media tool inside the Premiere timeline and an assistant into After Effects. Blackmagic shipped DaVinci Resolve 21.1 with a native MCP server. And a browser-based assistant editor called Eddie shipped a version that writes a real project file. Read together, the announcement set describes a single shift: the machine stopped being a generator bolted onto the pipeline and started being a participant in it.

The handoff artifact is what actually changed

Every automated editing tool has to hand something to the next person. For two decades that something was an interchange file: an XML, an AAF or an EDL, a list of cuts and timecodes that a different piece of software could reimport. An interchange file describes a decision. It does not preserve the workspace that produced it.

Eddie AI v4 does something different. It writes a native Premiere Pro project or a native Resolve project with bins, multicam angles and source timecode intact, ships an Avid extension, and delivers into a Premiere Productions or Blackmagic Cloud project. Its exports point at the original camera files rather than proxies. A real project file, as one reviewer of the release put it, is now what an assistant has to hand you, and any tool still handing over an XML is behind.

The distinction is not cosmetic. A project file arrives with structure: bins that map to a scene breakdown, angle stacks that can be re-cut, timecode that survives a relink. An interchange file arrives with a result. Either way the cut can still be changed, but only one of those handoffs lets a team change the intent rather than the timing.

There is a second thread worth noticing. Avid moved its OpenTimelineIO adapter to version 2.0 in the same release cycle, with the stated aim of improving fidelity when a cut is handed to Resolve or a visual effects house. Interchange did not disappear. It was demoted to the seam between finishing stages, while the assistant's handoff to the editor became a working project.

Agents took the preparation layer, not the cut

The more useful reading of IBC 2026 is where the agents were allowed to work. Avid's Ready to Edit handles sync, dialogue transcription and alignment, and bin organisation. Its embedded Gemini panel generates B-roll, finds stock footage and automates tagging, logging and translation. PhraseFindAI Multicam Auto Cut, shipped in Media Composer 2026.8, reads a transcript, works out who is talking, and builds an editable multicam sequence that switches angles on speech boundaries.

None of those is a creative decision. They are the hours before a creative decision: finding the take, matching it to the transcript, getting angles into a stack, naming the bins so the next person can find anything. Avid's own framing is explicit. AI is meant to speed up the editor's work without deciding for him, and narrative selection stays under editorial control.

The {{link}} problem also gets an answer here, because the barrier was never capability. It was that autonomous tooling with no permission model is unusable in a room that runs version control and client approvals. Avid Assist, the orchestration layer unveiled alongside the editing announcements, puts agents under permissions, policies and business controls and stays agnostic about which language model runs them. That is what production infrastructure looks like when it absorbs agents: less autonomy, more audit.

The browser-based Media Composer points the same way. Avid is not replacing the desktop editor. It is moving search, selection and preparation into a browser so that people who are not editors can do the front half of the job without booking an editing station.

The agentic AI barriers in creative problem also gets an answer here, because the barrier was never capability.

What the editors already doing this report back

Avid released research alongside the show that is more useful than most vendor data, because it surveys the people who sit in the chair. In a survey of professional editors, 79% reported using AI in their workflows, with speech-to-text and transcription the most common use case at 60%. That number should end the debate about whether editors accept this.

The next three percentages describe the gap. 87% agreed AI is most valuable when it helps them achieve their creative intentions more efficiently, 76% said its best use is unlocking more time for storytelling, and 63% said AI tools need tighter workflow integration to be genuinely useful. Read in sequence, those answers say one thing: editors are not short of capability, they are short of less friction between the capability and the timeline.

That demand is being met by plumbing rather than by features. The clearest signal at the show was how many vendors shipped an MCP server, the mechanism that lets an assistant call functions inside an application. Resolve Studio's version exposes transcription, multicam building, audio normalisation and render presets to a scripting API. A cloud storage vendor shipped one for its editing integration, and Eddie shipped one too. Three servers at one show is not a product trend. It is an interface becoming standard.

The {{link}} still applies, but the queue has moved deeper into the pipeline. The wait that used to sit in front of a render is now the wait in front of a preparation step, and this year that step is being automated rather than shortened.

The AI video render queue bottleneck still applies, but the queue has moved deeper into the pipeline.

Provenance moved into the export, and that changes who owns the record

One detail sits quietly inside the Eddie release and deserves more attention than it got. The tool signs its output with C2PA content credentials, the provenance standard behind Content Credentials, so a broadcaster can check where footage came from and what was done to it.

That relocates provenance from an administrative afterthought to an export setting. Until this year the provenance conversation usually happened after the cut: someone in legal or brand assurance asked what the model had been trained on, and the answer arrived too late to change anything. When the signature is generated at the moment of finishing, the record exists before the question does.

For a commercial team the practical consequence is that compliance stops being a review stage and becomes a pipeline setting. The verification question shifts from whether a finished film can be defended to whether the tool that produced it can be trusted to state what it did. That is a narrower question and a much faster one to answer.

What commercial video teams should take from IBC 2026

Three checks are worth running before the next AI-assisted edit starts. First, ask any tool that promises to prepare footage what it hands back at the end. A project file with bins and source timecode survives a change of mind. An interchange file survives only a change of timing.

Second, re-budget the human hour. If preparation is automated, the time concentrates at the approval gate, and the {{link}} pattern holds: the finishing stage is where AI settled because it is where judgement still has a price.

Third, put provenance on the export checklist next to loudness and delivery specs. It is now generated in the finishing stage, so it belongs with the other things a finishing stage is responsible for producing.

The broader lesson is about selection. With 63% of editors naming integration as the barrier, the advantage does not come from adding more AI tools. It comes from choosing fewer that write to the formats the pipeline already reads. The {{link}} describes what a finished commercial deliverable now involves; the editorial handoff is simply the newest item on that list.

If preparation is automated, the time concentrates at the approval gate, and the AI in the finishing trades pattern holds: the finishing stage is where AI settled because it is where judgement still has a price.

The the AI video delivery era describes what a finished commercial deliverable now involves; the editorial handoff is simply the newest item on that list.