You asked me to find out what's really out there in creative AI right now — the tools that don't just make one thing, but take a whole creative job (a video, a podcast, a week of social posts) and do it end to end, like a little factory. I went deep, then had a panel of AI critics tear the findings apart, then made the research defend itself, then wrote this. Below is the honest answer: what it is, why it matters to us, what I found, how it works, and what I think we should do.
The short answer — if you read nothing else
- The winning tools aren't the ones that just make a clip or a voice. They're the ones that do the whole job — write it, make it, check it, polish it, publish it. Making raw content is now cheap and ordinary; doing the whole job, on-brand is where the money is.
- Most "AI agents" talk is marketing. Almost everyone says their tool uses clever AI "agents" working on their own. In reality it's usually a fixed assembly line with a quality-checker bolted on. Real, self-running agent teams are rare — and the boldest one that tried ("the world's first AI marketing chief") failed and shut down.
- Our real advantage isn't the tech — it's something we already own. Anyone can plug these tools together. What nobody else has is BCG's brand rules and our senior creatives' taste — the record of what we approve and reject. That judgement is the hard-to-copy thing.
- So we should flip the obvious plan: sell our ideas and judgement as the premium product, and let AI cheaply mass-produce the variations underneath (all the sizes, languages and versions).
- Be realistic about cost and effort. Making lots of AI video is expensive (you pay by the second), and a trustworthy "brand-checker" is a real engineering project, not a weekend hack. Plan and budget for that.
01 · The context
Why this matters to us
Three things are happening at once, and they all point the same way:
- Making content is becoming cheap and ordinary. Anyone can now generate a decent image, voice or video from a sentence. When everyone can do it, that skill stops being special — and the value moves to whoever can do it on-brand, safely, and at scale.
- The work is moving toward studios like ours. About a third of big brands say they'll bring nearly all their creative work in-house within a year, and the giant ad agencies are shrinking. The budget is flowing toward exactly the kind of in-house team we are.
- There are real legal traps. Courts have signalled that work made purely by AI (with no real human author) may not be protectable as your property — and there are billion-dollar lawsuits over how these tools were trained. So "a human clearly led this" isn't just nice to have; it protects the work.
When making things gets cheap, judgement gets expensive.
02 · What I found
What's actually out there
Video
- The "whole job" tools — type an idea and they plan it, make the scenes, add a voice and edit it together: Runway, Google's Flow, HeyGen, LTX Studio.
- The steady money is in business video — tools that turn a script into a presenter-style training or explainer video: Synthesia (used by 9 in 10 of the world's biggest companies) and HeyGen.
- The big warning sign: the most hyped consumer video app of last year, OpenAI's Sora, was shut down this year — it cost too much to run. A reminder that "cool" doesn't pay the bills; useful, on-brand work does.
Podcasts & audio
- The standout trick: drop in a document and get back a finished, two-host podcast in one click. Google's NotebookLM made this famous; ElevenLabs, Wondercraft and others do it too.
- ElevenLabs has become the go-to for AI voices and is now worth about $11 billion — a sign of how much value sits in this layer.
Social posts & ads
- "One thing into many" is the killer use — feed in one long video or one master ad, get back dozens of clips, posts and local versions: OpusClip, Jasper, HubSpot.
- The honest standouts (the few that really do split the work into specialist steps): Jasper and HubSpot. Most others just look like it.
- The cautionary tale — "Icon": it promised a fully automatic "AI marketing chief" that did everything with no humans. It didn't work, quietly switched to using real people instead, and collapsed. Full auto-pilot isn't ready.
Images & design
- Adobe is furthest ahead at the "whole job, on-brand" idea, and — importantly — it legally protects you if you use its images commercially. That's the bar we'd be measured against.
- Best-in-class for specific jobs: Midjourney (look and feel), Ideogram (text in images), Recraft (logos and brand assets).
03 · How it works
How these products are built — in plain terms
Strip away the jargon and it's simple. A "creative AI product" is just a few AI tools wired together into a set of steps, with a checker and a human sign-off. Two words get thrown around a lot — here's what they actually mean:
Two terms, defined
A "workflow product" = one tool that does a whole multi-step job end to end, instead of just one step. (Brief → idea → make → check → publish.)
An "AI agent" = a little AI helper that does one step and decides how on its own, then hands off to the next. A team of these is a "multi-agent" system — powerful, but harder and pricier to run, and rarer than the marketing suggests.
The sensible way to build one is an assembly line you can trust: fixed, predictable steps for most of it, a smart AI "brand-checker" that flags anything off-brand, and a human who signs off before it goes out.
That smart brand-checker — the part that makes it ours — is the hardest, most expensive piece to build well, and it needs real engineering care to keep trustworthy. The good news: that difficulty is exactly why a competitor can't quickly copy it.
The big bill isn't the thinking — it's the making. AI video is charged by the second, so churning out a few hundred versions of a video can quietly run to thousands of dollars in one go. Any tool we build needs a spend limit and a human check before the expensive step, not just at the end.
04 · What we should do
The plan
- Bet on our judgement, not the tech. Build our edge on the one thing rivals can't copy — BCG's brand rules and our senior creatives' taste. The tools underneath are swappable plumbing.
- Sell the thinking; automate the grunt-work. Charge for original ideas and craft (the premium bit); use AI to cheaply mass-produce the variations underneath — every size, language and version.
- Buy the tools, build only the brand-safe "front door". Don't reinvent the generators (Adobe, Google, ElevenLabs already make those). Build the layer that keeps everything on-brand, legal, and human-approved.
- Keep a human in the loop — always. It protects the work legally, it keeps quality up, and "made with real people" is now something clients actively want.
- Only call it "agents" if it really is. For the June hackathon, show a demo where you can see the steps hand off and a brand-check catch a mistake. Don't dress up a single AI call as a "team of agents" — people see through it.
The first thing to build (good for the June 22 hackathon)
Take one approved master ad and automatically make all the local-market versions — every language, format and channel — with our brand-checker watching and a human signing off at the end.
- Why this one: it's exactly where our brand rules and BCG's trusted, careful approach beat a generic tool — and big regulated clients pay well for it.
- How: buy the generators and the wiring; build only our brand-check + sign-off layer.
- The path: hackathon demo → one real client → a repeat, paying engagement.
05 · How I made this (and what to trust)
The honest bit
My method
I ran five deep searches at once (video, audio, social, the under-the-hood tech, and the money), reading and cross-checking real sources. Then I had five independent AI critics attack the findings, and made the research defend itself and re-check every shaky fact. A few claims got corrected in that process — this page is only what survived.
- What's solid: the big picture, the Sora shutdown, the legal traps, and the under-the-hood cost facts are well sourced.
- What to take with a pinch of salt: some companies' exact revenue/value figures are estimates, and a few "agent" claims are really just company marketing — I've flagged those.
- What I didn't cover: gaming/3D, coding tools, and players outside the US and China.
The full long version, the critics' notes, and the sources all live in the project files (research/creative-ai-workflows-2026/) if you ever want to dig in.