The Drip
AI hot takes are sending mixed signals. Anthropic and OpenAI both announced massive consulting partnerships last week — Wall Street firms, Big Four — to help enterprises adopt AI. Same week, the "consulting is dead" takes were everywhere on LinkedIn. Pick one. Either AI does it all itself, or companies need help integrating it. Both can't be true. We worked through it on this week's episode.
Inside The Bottle
The headline says consulting is dead. Anthropic just signed a partnership with JP Morgan. OpenAI did the same with Big Four. So which is it?
Justin's read: there are two categories of AI work, and they're behaving very differently right now. Building stuff — websites, scripts, applications — has momentum. The technology is closer to deterministic and improvements compound. Talking with AI — skills that ask questions, planning workflows, agents that serve customers — is fragile. A model release can change the personality overnight. The exact same skill that worked Tuesday hallucinates Wednesday. If your business depends on the second category, model changes are now a customer experience risk.
Kellan's read: every major technology era promises self-service, and what people actually mean is "self-service from a technical person." The spreadsheet builder. The someone-who's-comfortable-with-computers. The person who can read what the AI built and figure out what's wrong with it. That role isn't going anywhere. If anything, the people who got cut "anticipating AI" probably should have been the ones trained to do this work instead. Consulting isn't dead — the shape of what it looks like is changing. The partnerships are the proof.
The displacement question gets uglier. AI is good enough now that some entry-level admin work is at risk. There are studies suggesting that for some of those roles, the cost of running the AI compute is actually higher than the employee. So you've broken the on-ramp without saving any money. Even if the math eventually works, you've eliminated the path that gets people to mid-level. There's a real opportunity for firms to figure out what the new ladder looks like — and most haven't tried yet.
The cautionary thread: AI moves fast. Faster than humans can manage agents at scale. There's an old IBM paper from the late 70s that said a machine can never be allowed to make a decision because you can't hold the machine accountable. That one's coming back around.
Bottom line: more humans, not fewer. Better trained on the technology, not ignored. The folks panicking about being replaced aren't being told the truth — and the folks who think AI handles everything haven't actually built anything.
Lab Notes
| ■ | Justin's note: Memory used to be the fix. Six months ago, the advice was "put your context in a knowledge base." Now the AI writes its own memory, and it's creating conflicts — memories overriding actual core knowledge, or anchoring on details from old sessions as if they're load-bearing. The fix is becoming the problem. |
| ■ | Kellan's note: AI creep is the new scope creep. It adds things you didn't ask for, and you didn't save any time — you moved the work to "fix this thing later." Beginning, middle, end. Not beginning, middle, middle, middle. Discipline is the most underrated skill. |
Latest Article
AI Pathfinder: Why Mapping Value Matters
"Can AI do this?" used to be the question. With generative AI the answer is nearly always yes — which is why mapping the value your work creates is the actual first move. Step 1 of the 5-step Pathfinder method.
What Stopped Our Scroll
| ■ | OpenAI's new consulting arm — DeployCo raised $4B at a $14B valuation from TPG, McKinsey, Bain, and Capgemini. The legacy firms just funded a front-row seat to their own disruption. |
| ■ | Google: AI-powered hacking is already here — Google's threat team caught hackers using an AI model to find and exploit a zero-day. The official framing went from "imminent" to "already begun" inside one report. |
| ■ | GitLab's Act 2, with a giant asterisk — GitLab is cutting country footprint 30% and doubling R&D pods on the bet that agentic engineering expands their market. Simon Willison notes their stock is down 50%, so they're financially motivated to believe exactly that. |
|