We’ve received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5.
We'll begin restoring access tomorrow, and will share an update soon.
We’re grateful to our users for their patience, and to everyone who worked with us on redeploying the models.
Programmatic doesn't have a performance problem. It has a measurement problem.
Click attribution can't see it, so the platform calls it dead.
The only honest read: hold spend in one matched market, keep it in another, measure the revenue gap.
Stopped asking AI to do tasks. Started building systems that run them.
The forecasting still takes back and forth. But each round I capture what works into its instructions.
It improves because I'm teaching it, not because it figures itself out.
Biggest AI lesson lately wasn't about prompting. It was sequencing. Lock the narrative first, then let AI handle the complex slides and skip the easy ones. Half the prompts, fewer revisions. AI gets useful once the thinking is done, not before.
One day into Prime Day, CPCs on a CPG brand we manage are running almost double last year. Discounts are shallower, traffic is up, margins are tighter. Demand isn't gone. It's just more expensive to capture. What are you seeing?
Platform ROAS said cut budget. MER said don't.
We kept investing in Meta TOF for a CPG DTC brand at 2.3x ROAS. MER was 5.1x. Now we're running an MMM to validate iROAS heading into Q4.
The metric you optimize for determines what you see.
Most people using Claude are stuck in phase 1.
Prompt. Review. Rewrite. Repeat. You are the bottleneck.
Phase 2: define the goal once. Agent executes and iterates.
Built for budget allocation. Runs itself after setup.
What phase are you in?
Slack message in. 75% complete strategy deck out.
Claude had the context loaded. Slack was just the trigger.
Most teams still manually prompt AI. The unlock is embedding it into how work flows.
Most Claude workflows fail before a new user ever touches them.
Test in incognito before you hand anything off.
Incognito strips memory and chat history. What's left is exactly what a new user sees.
Ghost icon. Required files only. Run the prompt verbatim.
Most people start every Claude chat from scratch.
Built a memory system that loads full context automatically on every chat.
Full walkthrough on YouTube: https://t.co/xZhfYZ3cGC
Most paid creative strategy skips straight to concepts. Hook ideas, formats, angles. All disconnected from the brand work that was supposed to inform them. Built a 4 skill Claude workflow that forces the connection. Assessment → Personas → Creative Direction → Testing Roadmap
Most people set up AI memory once. Never update it.
Three months later the forecast looks right. Built on January inputs. Wrong budget, wrong goals, wrong channel mix.
Fix: end of every session, one sentence per change. "Budget is now $120k. TikTok is now priority." Two minutes.
CPG budget planning used to mean quarterly spreadsheet rebuilds.
Built an AI workflow in Claude that answers:
How much budget to hit 15-20% penetration in target DMAs? Given a fixed budget, which DMAs do you prioritize?
Two scenarios as output. Leadership picks, not rebuilds.
Claude's memory problem isn't storage. It's structure.
Flat files don't show relationships. Obsidian does.
Building a linked vault where my context files connect to each other. The leverage is in the links, not the files.
Early days — but it's working.