Cutting through the noise
The agency conversation about AI has matured past the "it will replace us / it will save us" extremes. In 2026 the actual question is operational: where does generative AI quietly increase quality and throughput in a senior creative workflow, and where is it still expensive distraction? We use generative tools every single day at DIGISUBS — but the list of places they have replaced something is shorter than the list of places they have augmented something.
This post is our honest accounting. The tools change every quarter; the principles seem to be more durable.
What's working
Five workflows where AI has materially improved our quality of output:
Early ideation. When we are exploring brand directions, generating fifty image concepts in twenty minutes via Midjourney, Krea, or the latest Stable Diffusion fine-tune is genuinely useful. We do not ship any of those concepts directly — they almost always have a generative-AI fingerprint that an experienced eye reads as cheap — but they widen the search space we explore before we land on the direction we will hand-design. It is an idea machine, not a deliverable machine.
Copy iteration at scale. Writing twenty headline variations for an A/B test used to take a copywriter half a day. Now we write four, generate sixteen variants from a frontier model, edit them down to ten that don't sound like AI, and ship the test. The senior creative attention is on selection and editing, not on first-draft volume.
Code scaffolding. Most of our front-end engineers use Cursor, Claude Code, or a similar AI-native editor as their default. The productivity gain is real, especially for boilerplate-heavy tasks: setting up a new component file, writing a Storybook entry, generating types from a JSON schema. We have not seen AI replace senior engineering judgement, but it has made our junior engineers materially more productive.
Audio and motion drafting. Voiceover demos, draft music beds, rough animatics — generative tools are now good enough to produce placeholder assets that real production teams can use as creative briefs. Two years ago a director would describe what they wanted in words. Now they generate it, hand it to the production team, and say "make this, but better". The brief is sharper.
Document summarisation. Almost every long form document we receive from a client (RFPs, brand guidelines, customer interviews) gets summarised before we read it manually. The summary is rarely the final read, but it tells us what to look for in the deep read. It is a search index, not a substitute.
What we tried and quietly abandoned
Three workflows that sounded promising and did not pan out:
AI-generated final brand assets. We tried it. The results are detectable to a trained eye and uncanny to an untrained one. Logos, illustrations, and key brand assets still need to be hand-crafted. The reason is not snobbery — it is that brand assets are reused thousands of times across thousands of contexts, and the slight distortions of generative output compound across that surface area in ways that erode brand quality.
Auto-generated user research. We experimented with synthetic user interviews — feeding a persona spec into a frontier model and asking it to "respond as that persona". The output was plausible but eerily aligned with whatever assumption was already in the prompt. It is excellent for stress-testing a brief; it is dangerous if you treat it as research.
Full-pipeline campaign generation. A 2024-era pitch was that AI could generate an entire ad campaign — concept, copy, visuals, layout — end to end. The 2026 reality is that no campaign we have seen produced this way moved any meaningful business metric. The output averages well but never lands on a specific cultural moment, which is what good campaigns actually do.
The principle behind our taste
The pattern is consistent: AI is excellent for the middle of the creative process and weak at the ends. The beginning of a project — defining the problem, the audience, the strategic angle — is where senior creative judgement compounds and AI tends to be confidently wrong. The end of a project — the final assets that will live in the world — is where craft fingerprints matter and AI output is detectably synthetic.
The middle is where AI shines: generating volume, exploring variants, drafting placeholder assets, accelerating the journey between problem and solution. We treat it as an exoskeleton for the team. It does not replace senior judgement; it makes one senior judgement go further.
How we structure the team around it
Practically, we have made three changes to how the studio operates because of AI tools:
Tooling is part of onboarding. Every new hire spends their first week learning the team's AI workflow conventions — which tools to use for which tasks, what prompts work, what parts of the pipeline are AI-prohibited (final logo work, sensitive client copy). The AI workflow is part of the studio's craft now.
Senior reviewers, not prompt engineers. We do not have a "prompt engineer" role. We have senior creatives who use AI tools the way a chef uses a sous-vide — as one technique among many. The judgement is still the senior creative's; the tool just speeds up some of the steps.
Clear client communication. We tell every client up front: AI is part of our process, mostly in ideation and drafting, not in final asset production. We document exactly which deliverables involved which tools. Transparency removes the awkwardness of clients eventually discovering it through someone else.
Where we think it goes from here
The next twelve months are likely to bring video-generation tools that finally cross the quality threshold for marketing use, multi-modal models that can edit existing brand assets while preserving identity rules, and design copilots that understand design tokens natively. We are watching, but we are not building any client deliverables on top of unreleased capability. Tools that arrived with breathless announcements have a way of slipping their timelines, and a creative pipeline that depends on a not-yet-shipped tool is a pipeline that will let the client down.
The boring conclusion: AI is a quiet competitive advantage when applied with discipline, and a fast track to mediocrity when applied with hype. The studios that stay senior, taste-driven, and selective about the tool's place in the workflow will compound. The ones that try to ride the hype curve will produce a year of average work and lose the talent that produced their good work in the first place.