Good news: AI is growing up. Bad news: AI has moved from the kid’s menu to adult appetizer pricing (and entree costs are no doubt in our future).
The economics are catching up with the hype, the capabilities keep pushing past what most people expect, and a meaningful gap is opening between daily power users and everyone else.
Agency owners need to pay attention to all of it.
The end of AI’s free lunch
The uncertainty around Claude pricing is a sign of more cost turbulence to come. Simon Willison’s walkthrough captures how quickly the economics of AI have shifted, and how hard it is for even experienced users to predict what they will actually pay. Combined with Anthropic’s moves against OpenClaw earlier this month, the pattern is clear. Flat-rate all-you-can-eat pricing was never economically viable for agentic workloads, and the AI companies are correcting it in real-time.
Clients are getting more commitment-phobic. A SaaStr piece this week made the case that customers are pushing for shorter contracts across SaaS, not because they doubt the product but because they cannot confidently predict what the market will look like in 18 months. Three-year deals fell from 28% of new contracts in 2023 to 23% in 2026. Sub-one-year deals more than tripled, from 4% to 13%. While this is SaaS data, the same pressures face agency negotiations.
GitHub paused new sign-ups for paid Copilot plans this week. Microsoft-owned GitHub announced that it is pausing new Copilot Pro, Pro+, and Student subscriptions and tightening usage limits for existing customers. The company admitted it is now common for a handful of user requests to rack up costs that exceed a plan’s monthly price. Reporting suggests Copilot’s weekly operating costs nearly doubled since January, pushing Microsoft toward token-based billing.
Why it matters: Agency owners should expect the same dynamic in their own client conversations. Clients who are not sure exactly where things are headed see long commitments as a bet they do not want to make.
What should you do?
First, don’t price or structure your services around cheap AI assumptions. The costs behind your workflow tools are getting more expensive, and the day-one pricing you signed up for is not likely to last.
Second, focus on retention by providing real value rather than depending on contract lock-in. You need to make an impression and start generating noticeable results for clients in the first 90 days. If you’re banking on a fixed contract length to succeed, you’re in trouble.
The capability curve keeps climbing
Design capabilities continue to become more accessible to agencies. Anthropic launched Claude Design, a new Anthropic Labs product that generates polished visual work from conversational prompts and positions itself directly against Figma, Canva, and Adobe. Anthropic’s chief product officer resigned from Figma’s board a week before launch, and Figma’s stock dropped another 5% on the news.
OpenAI stepped up its agent game this week. Workspace Agents are Codex-powered agents that run autonomously across Slack, Google Drive, Salesforce, Notion, and other enterprise apps, handling multi-step workflows even when users are offline. They replace custom GPTs and will move to credit-based pricing after May 6. Google is working on its own desktop agent to compete with Anthropic’s Cowork.
Did the Mythos story get scarier? Anthropic confirmed this week that it is investigating a possible breach of Mythos, the cybersecurity model it released to a handful of trusted partners just two weeks ago. The unauthorized access reportedly came through a third-party vendor environment, not Anthropic’s own systems, but the incident underscores exactly why the company was cautious about releasing the model in the first place. The New York Times is continuing its coverage of the story, and federal officials along with leaders at key companies and institutions are paying close attention.
Why it matters: The AI capabilities that are now readily accessible to agencies create huge opportunities. The question is not whether to adopt. It is how to keep your evaluation current without getting bogged down in tracking and testing every fresh release.
At the same time, the Mythos story reminds us that AI progress comes with risk alongside rewards. Anything your team is doing with AI on client accounts deserves the same kind of careful thinking about access, governance, and vendor risk.
The widening perception gap
Andrej Karpathy crystallized something this week that a lot of agency owners are already feeling. In a widely-shared post, the former OpenAI co-founder argued that there is a growing gap in understanding of what AI can actually do. Many people tried the free tier of ChatGPT last year, saw it fumble basic questions, and formed a view. A much smaller group uses frontier agentic tools every day for complex technical work, and for them the experience is disorienting. These two groups, he argued, are talking about completely different products.
Dan Hockenmaier’s common-sense essay is the useful counterweight. His argument: some people have hit another gear with AI, but many others have gone backward by chasing every new tool, wasting time on “productivity theater,” and shipping slop that undermines the trust of colleagues and clients. The difference between the two groups is having a clear answer to which problems are actually worth applying AI to.
The labor market data supports both views. New UK research laid out the evidence so far on AI’s workplace impact, while OpenAI itself has started encouraging firms to trial four-day work weeks to adapt to the AI era. And Futurism reported on mounting signs of a consumer backlash, with concerns ranging from energy consumption to the quality of AI-generated content flooding online spaces.
Why it matters: For agency owners, the practical implication is that your clients, your team, and your prospects are not having the same AI conversation. Some are power users who expect AI-native workflows. Others formed their views 18 months ago and assume the technology is still clumsy and unreliable.
Your positioning and your internal training need to account for both. The worst place to be is stuck in the middle, assuming everyone has a sophisticated view of AI when half the room is still skeptical and the other half is five steps ahead of where you expected.
AI visibility continues to be a hot topic
The race to dominate AEO/GEO is on. Kevin Indig and Search Engine Journal both made the case this month that shorter, tightly focused content wins inside ChatGPT, a shift from the long-form SEO playbook that has dominated agency content strategy for the past decade. There is also new content circulating on practical tactics to make ChatGPT surface your business.
Why it matters: Not every agency needs a full AEO or GEO practice, and the hype in this space is running ahead of the reality. But there is no doubt that AI discovery is becoming a meaningful channel for buyer decisions, and the playbook may be different from traditional SEO.
The bottom line
The AI story for agency owners is not about whether to pay attention. It’s about paying attention to the right things.
Three key takeaways from this edition:
- Economics are tightening. Your service offering and pricing need to integrate more cost planning than they did a year ago.
- Capabilities keep climbing. Your clients and competitors may be using the latest tools if you’re not, so you need to invest in identifying, learning, and integrating the right ones for you and your team.
- The perception gap is widening. AI enthusiasts and skeptics seem to be coalescing in their own camps, and that creates challenges for agencies who need to work with each one – both internally and with clients.
