Agencies have bought into AI adoption. Everyone has a subscription. Everyone is prompting. Everyone is telling clients they are “leaning into AI.”
But there’s a difference between performative AI and AI that performs.
This week brought a cluster of stories that illuminate the gap: developers gaming usage dashboards rather than doing real work, an autonomous agent wiping out a company’s entire database in nine seconds, and a wave of AI-generated content so bland that audiences have stopped noticing it at all.
The agencies that will succeed are the ones treating AI as a tool that requires judgment, not a mandate that requires compliance.
The quality problem is getting harder to ignore
A lot of what passes for AI-assisted content right now is just a new flavor of an old problem. Writing in PR News, Kiri Jewell argues that “deck talk” is a bigger threat to thought leadership than AI-generated content.
Deck talk is what you get when people write the way they present: heavy on jargon, light on answers. A focus on performance and promotion rather than insight and ideas. Without enough editorial gatekeepers, it is more prevalent than ever.
The fix is not to use AI less. It is to make sure the end result is useful, unique, and genuinely engaging, regardless of how it was produced.
A related piece from Search Engine Land puts a name on what happens when you don’t: the “bland tax.” Semrush CMO Andrew Warden argues that AI systems are conditioning themselves to ignore sameness. Generic content does not just fail to stand out. It gets mashed up into a single synthesized answer with no attribution, or filtered out entirely. Your content becomes invisible in AI-generated responses without anyone noticing the cause.
The advertising world is feeling this acutely. Rachel Lyndon-Jones argues in The Drum that what audiences are experiencing is not ad fatigue but cultural exhaustion from undifferentiated AI-generated content with no real point of view. The value of genuine agency creativity only increases as lazy AI output floods the landscape.
And it is not just humans you need to reach. Elena Verna, writing in her Substack, makes the case that AI agents are increasingly the ones interacting with your clients’ products and content, often without a human ever touching them. Communicators need to understand how to talk to machines while also connecting with real people.
Why it matters: The bar for content that earns attention from humans or AI systems is rising faster than most agencies are adjusting for. Differentiated thinking from real experts is more valuable now than ever. Agencies producing volume without distinction are paying the bland tax whether they know it or not.
When AI mandates backfire
Inside Meta, developers were competing on an internal leaderboard ranking employees by AI tokens consumed, not by what they produced. The Pragmatic Engineer reported on “tokenmaxxing,” the practice of deliberately inflating AI usage to look good on internal dashboards. Meta employees burned 60.2 trillion tokens in a single month, roughly $900 million at Anthropic’s API pricing, with engineers loading entire codebases as context and spinning up agents on tasks nobody intended to complete. Meta eventually took the leaderboard down.
The dynamic was fueled by Nvidia CEO Jensen Huang’s comment that if a $500,000 engineer did not consume at least $250,000 worth of tokens, he would be “deeply alarmed.” When leadership signals that usage is the metric, people optimize for usage.
The human cost of getting this wrong is real. A survey of 81,000 workers found that respondents experiencing the largest speed gains from AI also express the highest concern about job displacement. The people on your team getting the most done with AI may also be the most worried about what it means for their careers. Agency owners ramping up AI usage without acknowledging that anxiety are creating a retention problem.
And then there is what happens when AI is given too much autonomy without enough oversight. Jer Crane, founder of PocketOS, published an account of how a Cursor agent running Anthropic’s Claude deleted his entire production database and all backups in nine seconds. The agent hit a credential mismatch during a routine task, decided on its own to fix it by deleting a storage volume, and made a single API call. No confirmation requested.
Why it matters: Blanket directives to your team to “use AI more” will produce the same result at a smaller scale. Never use AI for its own sake. The question is never how much AI your team is using. It is whether the work is better, faster, or cheaper than it would have been without it.
And to protect yourself and your clients, before deploying an AI agent on any task, ask one question: what is the worst thing this agent could do if it makes a wrong assumption? If the answer is catastrophic, put a human checkpoint between the agent and the action.
Cost/price pressures continue to rise
The cost story keeps getting more concrete. The Wall Street Journal reported that OpenAI missed multiple monthly revenue targets in early 2026 after losing ground to Anthropic, and fell short of an internal goal to reach one billion weekly active users. CFO Sarah Friar has reportedly warned colleagues she worries the company may not be able to fund future computing contracts if growth does not accelerate.
Meanwhile, GitHub announced all Copilot plans move to token-based billing on June 1, 2026, with a related report confirming Microsoft’s broader Copilot plans are heading in the same direction. Ed Zitron’s latest piece adds context: Uber burned through its entire 2026 AI budget in a matter of months, and Goldman Sachs data suggests some companies are already spending 10% of total headcount costs on AI tokens.
Why it matters: The point for agency owners is not that AI is getting more expensive. It is that AI must get more expensive to survive, and understanding the cost-value equation now puts you ahead of clients who will be asking the same questions soon. Too many agencies worry about jumping from $20 to $200 per month without doing the math on what they are actually getting. If the value is there, pay for it. If it is not, do something different.
What’s actually working
Adobe’s entire Creative Cloud suite is now accessible directly inside Claude. The Adobe for creativity connector gives Claude access to more than 50 tools across Photoshop, Illustrator, Premiere, Lightroom, InDesign, Express, Firefly, and Stock. Describe what you want, and Claude orchestrates multi-step workflows without switching apps. It is becoming increasingly possible to integrate AI across the tools agencies already use, rather than running AI in a separate workflow alongside them.
The adoption data is also showing real patterns. A Marketing Charts analysis finds that marketing productivity is up and overhead costs are down, but marketers are still relying on AI primarily for content creation. That means there is a long runway to apply AI to reporting, research, briefing, and planning, where the ROI case is equally strong but adoption has been slower.
Why it matters: The tools are getting genuinely better and more integrated. The agencies making the most of it are not the ones using AI the most. They’re the ones being deliberate about where it adds real value and adjusting as it evolves.
The bottom line
AI is accessible to every agency. Judgment about where and how to use it is not evenly distributed, and the gap is becoming visible in the quality of work agencies produce.
Three things to take away from this edition:
- Distinctive work is your competitive advantage. Poorly-managed AI flooding the market with generic content makes original thinking more valuable, not less. If your agency produces real expertise with a genuine point of view, you are more differentiated than you have been in years. If you are producing volume without distinction, you are paying the bland tax.
- “Use more AI” is not a strategy. Blanket mandates produce gaming, anxiety, and risk. Define what good AI use looks like in your agency, measure outcomes rather than activity, and make sure your team understands that their job is judgment, not prompting.
- Understand your cost-value equation. AI pricing is adjusting and will continue to do so. Get clear on where you are getting real value and where you are just spending money. That clarity will serve you when clients start asking the same questions.
