AI Cuts Creative Teams: What Companies Must Learn About AI Leadership
UK creative agencies reported a 14% decline in staff, with AI named as a factor. If work is becoming promptable, what's the role of a designer, a creative director, a product manager? The answer isn't output. It's judgment.
Platforms now generate and auto-optimize creative by default. When machines can ship endless content, the scarce ingredient is taste - the human ability to choose what resonates before data can prove it.
AI Is Average - And That's The Problem
Most AI optimizes for what's statistically likely. It's a machine for plausibility, not intention. That's fine for production, but "on average" is where differentiation goes to die.
The work that moves people rarely lives at the center. It lives on the edges: oddly specific, emotionally precise, strategically brave. That edge is taste.
What Happens When Platforms Pick Your Creative
In late 2025, a widely shared anecdote in ad circles described a platform's auto-creative swapping a top ad for an off-brand image - an older woman used to sell a men's product. It spread because it wasn't slightly wrong; it was clearly wrong.
That's the risk. Systems optimized for averages will drift toward what "usually works," not what your brand believes. Without taste-led guardrails, your differentiation gets flattened.
We Need A New Kind Of Creative Leadership
Leaders must learn to steer probabilistic systems, defend a point of view, and say no to the statistically plausible option. Call it taste leadership.
Taste is the filter that keeps "looks fine" from shipping. It protects the idea, the voice, and the edge that makes your work stick.
No, Creative Roles Aren't Dead - They're Changing
Designers, writers, product managers, and creative directors aren't going away. Their jobs are shifting from making every asset to directing systems, setting constraints, and curating what ships.
The highest value work is now selection, sequencing, and strategy - deciding what to make and what to reject.
A Practical Playbook For Creatives
- Write briefs for edges, not averages: State what the work must do, what it must avoid, and what "wrong but right" looks like. Include anti-goals and non-negotiables.
- Build a brand taste repo: Collect examples that are "us" and "not us," with notes. Turn it into style tokens, palettes, casting rules, and negative prompts.
- Human-in-the-loop QA: Red-team outputs for tone, audience fit, and cultural risk. Run pre-mortems: "What would make this blow up in the wrong way?"
- Measure beyond easy clicks: Track saves, replies, branded search, repeat view-through, price premium tolerance, and share-of-search. Optimize for signals your brand actually values.
- Constrain the system: Lock critical attributes (persona, age range, color system, claims). Turn off auto-creative swaps when brand risk is high.
- Test on the edges: Use micro-audiences, dark posts, and holdouts. Keep a human veto even if an early metric looks good.
- Tune with taste: Fine-tune on brand-safe sets and score outputs with human taste raters. Reward distinctiveness, penalize drift.
- Version everything: Keep a changelog for prompts, seeds, and settings. If a campaign goes sideways, you need to roll back fast.
- Governance beats chaos: Create a creative charter, escalation paths, and a kill switch. Decide who can approve boundary-pushing work.
- Protect your career: Ship case studies that show your taste. Keep a swipe file. Run weekly critique sessions with clear acceptance criteria.
Metrics That Protect Taste
CTR and CPC reward lowest-common-denominator bait. Use layered metrics that align to brand outcomes.
- Depth: dwell time, scroll depth, completion rate
- Intent: branded search, direct type-in, add-to-saves
- Advocacy: replies, quality comments, shares-to-saves ratio
- Durability: fatigue curves, repeat exposure lift, cohort retention
- Pricing power: willingness-to-pay deltas in tests
What To Automate vs. Keep Human
- Automate: Variants, resizes, translations, channel formatting, first-draft copy, alt text, concept thumbnails.
- Keep human: Core concept, headline logic, narrative arc, casting, color and type systems, final selects, and brand claims.
Working With Platform AIs Without Losing The Plot
- Feed context: Upload brand kits, voice rules, do/don't lists, reference reels.
- Set hard rails: Age/gender/setting constraints, compliance limits, claims guardrails.
- Close loops: Down-rank off-brand outputs, blocklist failures, and send feedback to your reps.
Taste Is Trainable
Taste isn't magic; it's reps and reference. Build a daily practice of observation, remixing, and critique. Ship small, learn fast, document what "clicked."
If you want structured reps with modern tools, explore role-specific programs and prompt workflows such as the AI Learning Path for Business Unit Managers, the AI Learning Path for CIOs, or the AI for Executives & Strategy.
The Bottom Line
Output is cheap. Taste is scarce.
AI will keep producing the average. Your job is to decide what's worth making - and protect it long enough for the market to feel it.
Your membership also unlocks:
