Adobe Wants Agents Editing the Original File

Adobe's planned integrations will let third-party AI agents edit native Photoshop, Illustrator, Premiere and After Effects files while keeping their elements editable. Alongside Instagram's Mac editor, Google's reference-based video generation and Ogilvy's licensed-content system, the news points towards creative tools that extend original work rather than continually replace it.

A campaign rarely stays finished for long. A price changes, a product sells out, a new format appears, or a client asks for a slightly different cut. This week's most useful creative announcements pay attention to what happens after the attractive first version has been exported.

A finished image is a dead end

Adobe used MAX on 10 November to preview local integrations based on the Model Context Protocol, or MCP, for Photoshop, Illustrator, Premiere and After Effects. The idea is that a compatible external AI agent can work directly with the files a designer already uses, while layers, vectors, bins and video sequences remain editable. Adobe says those integrations are coming to Creative Cloud customers in early 2027, so this is an announced direction rather than something all subscribers can use today.¹² That timing matters, because a preview can be strategically interesting without yet solving anybody's production backlog.

Consider a retailer whose winter campaign has been built as a layered design, with a product image, promotional badge, prices and text in several sizes. Two weeks later, the offer changes and the same creative needs to run in a narrower placement. If software returns a flattened picture, someone has to reconstruct the type, remove the old badge and possibly recreate the composition. The valuable file is the one you can revise, because it carries decisions that are expensive to reconstruct from the final image alone.

Adobe's example is deceptively ordinary: remove a logo from campaign assets, then ask an agent to carry that alteration across the related work. This is more credible than the usual demonstration in which a prompt produces a glamorous image with no messy follow-up. It also raises a useful test for any creative automation tool: can the person receiving the output change one detail without disturbing everything else? An impressive export may be a poor deliverable when the campaign has another six weeks to run.

Some of Adobe's other MAX updates reinforce this focus on continuing an edit. Premiere's Generative Media tools bring video, music and sound generation into the existing timeline, while the new Generative Remove beta targets unwanted elements inside footage. Photoshop's updated Remove Tool takes account of related visual details such as reflections and shadows, and Firefly Graph is in beta for repeatable workflows.¹ None of those features proves that a complex campaign can be changed reliably without checking the result, but each reduces the need to abandon the project and begin elsewhere.

For larger marketing teams, retaining a native file means retaining the ability to hand work to the next designer, the next agency or the same team next month. For smaller firms, the principle is identical even if they never open Photoshop. An export is a delivery format; an editable project is working capital for the next round of content. The most useful creative tool is increasingly one that knows the difference.

Instagram gets a working timeline

Instagram's Edits app arrived on Mac on 5 November with a multi-track timeline, colour grading, audio controls, several aspect ratios and watermark-free 4K exports. More unusually, its MCP interface lets a compatible AI agent use editing tools to assemble or change a cut, after which a person can carry on in the same editor.³ The rough cut becomes an editable starting point rather than an isolated video generated elsewhere. For anyone dealing with dozens of short clips, that is a material change in how an Instagram content strategy might be executed.

Picture the footage from a small restaurant's lunch service: the oven opening, plates arriving at the pass, a chef finishing a dish and the room beginning to fill. An agent could be asked to gather a short sequence around those existing clips, leaving the owner or marketer to decide whether the pace and sound resemble the real place. The business would still need to choose which moment tells the story. But it could spend less of its limited time placing clips in order and making routine adjustments.

There is an important distinction between an agent editing footage and a model inventing footage. A real kitchen recording has incidental details that contribute to a restaurant's identity, including the way staff work and how the space actually looks. A generated video may be useful in other contexts, but it cannot supply documentary evidence of what happened in that kitchen. For local brands, using their own photographs and recordings often does more for recognisability than adding one more visually polished piece of Instagram content generation.

The Edits announcement also has practical limits. The desktop product is currently for macOS, and projects cannot yet travel freely between its desktop and mobile versions, although saved audio and text styles can transfer.³ A team that films on phones and finishes on laptops should check the actual hand-off before promising a new weekly production routine. The convenience of an agent matters much less if the footage cannot move through the rest of the workflow.

This is where the category of Instagram AI content management could become more interesting than simple caption generation. Selecting footage, organising versions, adapting framing and keeping access to the original material are recurring jobs. They also generate real frustrations for the owner of a boutique, café or small product brand that has to post consistently. Software that removes those frustrations without flattening the business's identity has a useful job to do.

Google has a product-shape problem

Google's Gemini Omni 1.2 Flash update, announced on 5 November, approaches reuse from a different direction. The model can produce and edit clips up to 15 seconds, compared with ten seconds previously, and Google says the 720p generation price has fallen by 30% to $0.07 per second. It also reports better adherence to reference photographs, short clips and audio cues, including details such as garments, packaging typography and brand colours.⁴ Those are Google's claims about its model, rather than evidence that every output preserves every detail.

The distinctive point is what Google has chosen to improve. A fashion retailer cannot use a glossy video if the jacket's buttons shift position, the hem changes shape or the logo becomes a plausible-looking invention. An ecommerce team may spend less time fixing a clip if the model consistently follows the actual product images it has been given. For a business selling a physical object, accurate detail is part of the offer, not an optional aesthetic preference.

It is tempting to frame the lower price as the main development, because cents per second make a tidy headline. But generation price is only one cost in producing a usable advertisement. Somebody must select source images, write a brief, check the product depiction, confirm the text and make the final asset fit its channel. If a cheap clip demands three rounds of correction, its nominal generation price tells the team very little about the cost of publication.

Google has added conversational editing so users can request changes to a clip rather than regenerating it from the beginning. That is useful, although changing a generated video through prompts is not the same as opening a fully editable Premiere project. With a native project, a text layer or cut can be altered directly. With a generated clip, the model is being asked to produce a revised result that still needs inspection.⁴ Marketers buying these tools should understand which sort of editability they are receiving.

For an online shop, the immediate opportunity is modest and concrete. Start with photographs of the real product, use them to experiment with motion or short demonstrations, and compare the output against the item a customer can actually buy. Keep any original clips and product photographs rather than treating the generated variation as the new master asset. Reference material is the anchor, especially as models get cheaper and the number of versions becomes easier to multiply.

There is a useful consequence for creative budgets. If an existing product shoot can support a broader set of formats, the next pound might be better spent photographing the product carefully than requesting another batch of speculative renders. A good source image can help a person understand the item, support an accurate product listing and guide several later video experiments. The value of the original photograph grows when more tools can use it without losing what made it accurate.

Ogilvy cannot license a guess

Ogilvy's new Echo platform takes this question into celebrity advertising, where the original material carries contractual obligations as well as creative value. Announced on 9 November, Echo uses licensed digital likenesses through a partnership involving Creative Artists Agency and the CAA Vault. The system is designed to extend material from an existing shoot into more advertising assets while checking talent usage terms. Cleared consumer-facing recreations are marked to show that the likeness was approved and recreated using AI.⁵⁶

Ogilvy says a conventional shoot yielding eight to ten assets could support approximately 30 to 50 with Echo, and it estimates that average cost per asset might fall by 65% to 75%. Those figures describe the company's proposition, not independently verified results across real campaigns. Marketing Dive reports that pilots are in progress and that Ogilvy has not identified the brands or talent involved.⁶ The commercial case is plausible, but a marketer should not mistake a projected saving for the result of a completed campaign.

The more interesting feature is the connection between the asset and the right to use it. A performance can be recreated only within the permissions granted by the person whose likeness is involved. The same celebrity might permit one market, a specific product range or a limited campaign period without agreeing to unlimited appearances in every future advertisement. Storing a high-quality scan is not the same as owning unrestricted permission to use it.

This matters beyond famous faces. Smaller teams also reuse photography, customer testimonials, creator collaborations and music under arrangements that may be narrower than they assume. A model can make a variation technically possible without making it contractually or ethically acceptable. If the material contains a recognisable person, the production process needs to preserve what that person actually authorised. When a file is reused across formats, the terms governing it must travel with the file.

Echo's visible markings offer a further contrast with tools designed simply to make the synthetic material indistinguishable from reality. A celebrity recreation should not quietly pass as a newly recorded performance, particularly where the person never attended the supposed shoot. Brands that want the benefits of synthetic production have to make the relationship between original performance and later asset intelligible. That obligation is easier to meet when the source, permissions and production history remain attached to the work.

It would be a mistake to assume that a small business needs Ogilvy's infrastructure. A local salon can get plenty of value from a properly labelled folder of photos, written permission from people featured in a campaign and a record of where licensed music may be used. The lesson is about what the asset brings with it: details, context and rights that a prompt cannot safely invent. More creative output is useful only when the business can stand behind what it shows.

Keep the file that can change

These four announcements describe different products at different stages of availability. Adobe's local third-party agent integrations are a preview for early 2027; Instagram's Mac editor is available now; Google's updated video model is generally available; and Ogilvy is beginning to take Echo into commercial use.¹³⁴⁶ Their common commercial promise is not that every business will suddenly produce fifty pieces of content a day. It is that one piece of original material can survive more changes, more formats and more use cases before it has to be recreated.

A marketing team can act on that idea without buying anything new. Preserve original photographs and clips at full quality, keep editable design files where they exist, identify which product or campaign each asset belongs to, and retain the relevant usage permissions. Build the brief from details the business can prove, such as the product's appearance, the actual service or a recognisable moment from its working day. Then evaluate creative tools on what happens when the second request arrives: change the price, crop the format, remove a discontinued product or prepare another cut.

This should change how teams judge creative AI products. A demonstration usually shows the first result because first results are easy to admire. The second, third and fourth revisions reveal whether the tool has saved labour or simply moved it to somebody else. Ask to see a real campaign change made inside the same project, with its source material intact and its product details still correct. That test is worth more than a gallery of polished examples.

The original file used to look like something the production team needed to archive. It is becoming something marketing teams need to invest in, organise and protect. When software can do more with it, the quality of what a business has actually made, photographed and earned permission to use becomes more valuable. Keep the original. The next edit is probably already coming.

Sources