Last week, in our April 1st newsletter, I said I wasn't going to do an April Fools' joke. Reality beat me to it by a day.
On March 31, Anthropic accidentally shipped 512,000 lines of Claude Code's source code to a public registry. Developers found a Tamagotchi pet with RPG stats. A regex that logs every time you swear at the AI. An entire subsystem called Undercover Mode built to prevent Claude from leaking secrets shipped inside the leak itself.
The deeper story was the orchestration layer underneath all of it. Memory management, multi-agent coordination, background processing. The model was never the product. The system around it was.
But what stuck with me most is simpler than that. A company with $19 billion in revenue and an engineering team built to prevent this exact scenario got undone by a missing line in a config file.
The smartest system in the room still couldn't adapt to a basic human error. That's not a technology problem. That's an organizational one.
Which brings me to this week. As I write this, we're opening the Level Executive Summit here in Pittsburgh. The theme is Adaptive Advantage, how organizations improvise, compete, and win when the ground keeps shifting underneath them.
The agenda was built around that idea.:
Dr. Aniket Kittur from Carnegie Mellon is showing how AI can surface unexpected analogies that unlock innovation, the kind of lateral thinking no amount of automation produces on its own.
Google's Alana Parsons and Chris Marino are breaking down how measurement moves from signal to causal proof across a full funnel, which matters a lot when the channels keep fragmenting.
Reddit's Carly Steuck is here to talk about how buying decisions now start in communities, not campaigns.
And tomorrow night, Second City is running an improv workshop for our group.
The premise is that the ability to respond to what's actually happening beats any amount of scripted preparation.
Anthropic had the script. It didn't matter.
The next three days are about building the muscle that does. I'll share what comes out of it.
Patrick Patterson
CEO at Level Agency
What's interesting this week
Engineering AI-readable authority: a scalable GEO writing system Level Agency and Ancora Education are Shorty Award finalists in the GEO category for a writing system built around one core idea: stop optimizing for rankings and start engineering for AI retrieval. One optimized asset drove a 1,023% increase in clicks and became the second most visited page on the site. P.S. We'd love your support, check out our entry and vote for us in the GEO category. Read more
Seven independent agencies earn The Trade Desk's second Premier Partner class The Trade Desk named its second Premier Partner cohort this week, seven independent agencies certified across trader, marketer, and executive tracks through Edge Academy. We're proud to be among them. For our clients, it's a signal that the people running their programmatic campaigns are operating at the highest level of platform fluency The Trade Desk certifies. Read more
7 marketing lessons from Google's Media Lab Google's internal marketing team shared what they've learned running their own campaigns under an AI-first model. The framing that stuck with me: a marketer's job is to provide better fuel for the algorithm, not steer it. There's also a candid section on change management being the hardest part of AI transformation, which will feel familiar to most teams right now. Read more
As CFOs scrutinize CTV spend, incrementality emerges as a differentiator CTV budgets are under more scrutiny than they were a year ago. The agencies winning those conversations are the ones who can show actual lift, and incrementality testing has gone from a nice-to-have to a cost of admission pretty quickly. Read more
Really, you made this without AI? Prove it. A growing number of clients and platforms are starting to ask for proof that creative work is human-made, and a small industry is forming around verification and labeling. Whether this becomes a meaningful differentiator or just another box to check is still an open question, but the conversation is real. Read more
Deep dive: AI
Aware, uncomfortable, and still not buying: how consumers really feel about shopping in ChatGPT and Google
81% of shoppers in this survey knew in-platform purchasing existed. Only 27% have ever done it. 80% still prefer to buy from a brand's website, and that number hasn't moved in a year. This is a group of active, digitally fluent shoppers, so the resistance is worth taking seriously.
The most tech-savvy respondents are the most skeptical. They understand how recommendation algorithms work, and that familiarity is exactly what's driving their reluctance to complete a purchase through one. The concern showing up most in the data is commission-driven recommendations, the feeling that platforms are surfacing products based on what they get paid, not what's actually good.
One purchase changes everything, though. Shoppers who have bought in-platform once are dramatically more open across every channel. The barrier is getting someone to cross the line once, and a better checkout flow won't reach people who've already decided the answer is no.
The customer journey now centers on exposure, recall, and return
The attribution model most teams are still running assumes people search, click, and convert in a way you can track. AI overviews and aggregated answers are absorbing the informational layer of search, which means the click you see in your analytics is often the final step of a journey that started somewhere you had no visibility into.
The reframe in this piece is useful: exposure, recall, return. Your brand enters someone's awareness without a trackable interaction. Mental availability builds over time. Eventually, they seek you out. By the time the click shows up, the work that earned it already happened well before anyone landed on your site.
Branded search volume, direct traffic, and share of voice in AI outputs are becoming your leading indicators. The marketers who figure out how to measure what's happening earlier in that cycle are the ones who will actually get credit for the work they're doing.
Most AI tools live in their own tab. Sider lives in yours. It’s a browser sidebar that gives you access to GPT-4o, Claude, Gemini, DeepSeek, and Grok from a single interface, on top of whatever page you’re already looking at.
The practical value shows up immediately. Highlight a paragraph on a competitor’s page and get an instant rewrite or analysis. Drop a PDF into the sidebar and ask questions about it. Summarize a long YouTube video before deciding whether it’s worth your time. The group chat feature lets you run the same prompt through multiple models at once and compare answers side by side, which is useful when you want to pressure-test a positioning statement or see how different models interpret the same data.
Where it’s earned real attention is the Deep Research Agent. Give it a topic, and it scans over 100 sources and returns a structured report and a first-pass research brief that used to take half a day. Everything it finds gets saved to a personal knowledge base called Wisebase that compounds over time.
Why I recommend it: if your team spends most of the day in a browser researching, writing, comparing, or analyzing, Sider removes the friction of switching between AI tools and keeps the work in context.
What does ChatGPT actually say about you?
When a potential buyer asks an AI agent for a Top 5 list in your category, do you make the cut? More importantly: Why or why not?
Don't leave your brand’s reputation to an algorithm you haven't audited. Our scorecard pulls back the curtain on how AI platforms frame your brand sentiment and which sources they use to vet your authority.
Inside your custom report:
Sentiment analysis: Is the AI's summary of your brand accurate or outdated?
Citation audit: Discover which third-party sites are feeding the AI your info.
AEO roadmap: Immediate steps to improve your "AI discoverability."
