Making your own whiskey at home just became legal.
Yes, you heard me right.
A federal judge in Texas has declared a 156-year-old ban on at-home distilling unconstitutional.
The case is Hobby Distillers Association v. Alcohol and Tobacco Tax and Trade Bureau, U.S. District Court for the Northern District of Texas, No. 4:23-cv-01221.
Key points:
→ The ban was found to exceed Congress's taxing power
→ It also violates the U.S. Constitution's Commerce Clause
→ A permanent injunction bars enforcement against the association's members
The ruling is stayed for 14 days pending potential appeal.
This decision could be a game-changer for hobbyists looking to legally produce whiskey and bourbon at home.
The judge argued the ban:
→ Doesn't raise revenue, so it's not a valid exercise of taxing power
→ Can't be sustained under Congress' power to regulate interstate commerce
Is at home distilling going to be the next big trend?
Have seen some questions about the updated classifiers and wanted to clarify.
As with the original classifiers, a small fraction of routine coding and debugging tasks will be flagged and fall back to Opus.
We're excited for guys to get access back tomorrow.
Claude Fable 5 will be available again globally tomorrow.
After a series of productive conversations with the US government, we're redeploying the model with a new set of classifiers to target and block more cybersecurity tasks. In the near term, some routine tasks like coding and debugging will fall back to Opus 4.8. We’ll continue to refine these classifiers over the coming weeks to reduce false positives and better distinguish genuine misuse from legitimate requests.
We’ve also begun drafting a consensus framework—with Amazon, Microsoft, Google, and other Glasswing partners—for assessing the severity of AI jailbreaks and how AI developers should respond to them. We invite other industry partners and model providers to join us in this effort.
Finally, we’re scaling up our collaboration with the US government on model testing and safeguards. This will include pre-release access to models and safeguards for evaluation, information sharing on jailbreaks and misuse, and dedicated resources for joint research.
Thank you to our users for your patience, and to our partners across the government, industry, and the research community who worked alongside us to make Fable 5 available again.
Read our full blog: https://t.co/VHyum831ri
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.
we need to talk about sonnet 5...
composer 2.5 is faster, glm 2.5 is cheaper, opus 4.8 output is 10x better.
its an improvement yes. But without a real use case it is dead on arrival right now.
will you guys be running it?
Over the past two weeks, we have worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the US Government and strengthen America’s leadership in AI.
Since June 12, we’ve been working closely with the US government to restore access to Claude Mythos 5 and Fable 5. Today, the government notified us that Mythos 5, our strongest cybersecurity model, can be redeployed to a set of US organizations that operate and defend critical infrastructure.
We’re restoring access for these organizations quickly, and we’re continuing to work with the government to expand access to Mythos 5 and make Fable 5 available for general use again.
I Built a System That Runs 100 Deals a Week for a Firm I Advise (Steal the Pattern)
I recently built a deal-flow system for a firm I was advising.
Every week they get 100 deals in the door. Now an agent runs all of them, and a human only opens the ones worth their time.
But the best part isn't the deals. It's the pattern underneath. You can lift it out and run almost any repeated process through it.
Hiring, support tickets, content review, lead qualification, contract review.
Here's how I built it.
It runs on plain folders and a few skills. I'll use their due diligence flow as the example. Swap in your own workflow as you read.
Step 1: Pick a process you do over and over.
The whole trick is finding something repetitive with a clear "did it pass or not" at the end.
For this firm it's due diligence. Every week 100 deals come in, and each one has to run through a process unique to their firm to decide one thing: is this worth a human checking to see if it's worth investing in?
That's the job we handed to the agents.
Step 2: Build the folder structure.
This is the backbone. Inside the main "Due Diligence" folder, I created subfolders for each stage a deal can be in:
Due Diligence/
├ Pending/ ← new deals land here
├ Needs Review/ ← a human should review
├ Approved/ ← Ready for a human to proceed with
├ Denied/ ← didn't make the cut
This looks too simple to matter. It's the most important part.
The folders are a state machine your agent can actually see and move things through. Every deal lives in exactly one folder, and that folder is its status.
It also means the agent never reprocesses a deal it already judged. Anything sitting in Approved or Denied is done. Only Pending is live work.
Step 3: Feed the pipeline automatically.
A pipeline is useless if you have to fill it by hand.
So I set up a scheduled task that runs twice a day, once in the morning and once in the afternoon. Pick whatever cadence you want. That task searches the firm's email for any pitches that came in and /or pulls new entries from the database they use to collect them.
Every new deal it finds gets dropped into Pending. Now the front of the pipeline fills itself while the team does other work.
Step 4: The secret sauce.
Build firm specific skills and evals that run on each deal.
This is where your firm's edge lives.
For every deal that lands in Pending, we run a series of small, single-purpose skills:
The checklist eval. They have a 43-step checklist they use to evaluate a deal. The agent runs the deal through every step, searches the internet for any extra info it needs, and if it still can't make the call, it emails the founder directly to ask for what's missing.
The fit check. Does this deal fall inside the industries they focus on? Is it in a vertical they actually think is worth investing in? That's its own skill, with its own pass/fail logic.
You can stack as many of these as your process needs.
Keep each one small and focused on a single question.
When one skill is doing one job, you can improve it without breaking the others.
Step 5: Chain them into one command.
Once the skills exist, I chain them together with one orchestrator skill (or a plugin chain).
So when I say "run due diligence," it:
- Picks up every deal sitting in Pending
- Runs each one through the skills in parallel
- Reads the results and decides where the deal belongs
- Moves the deal into Approved, Needs Review, or Denied
The whole queue gets processed and sorted multiple times per day.
Tip: have it drop a short verdict.md inside each deal folder explaining why it landed where it did. Now you have an audit trail for every decision, and you can sanity-check the agent's reasoning in seconds.
Step 6: Visualize the pipeline.
I had Cowork build an HTML dashboard (a live artifact) that reads every folder and lays the data out visually.
One glance shows the whole pipe: how many deals are pending, what's approved, what's waiting on a human. The team can see exactly where every deal is without opening a single file.
Step 7: Run it on demand.
The schedule handles the routine flow. But the team isn't locked into it.
Any time, they can drag a deal straight into Pending and run the skill on the spot.
The real unlock: ship it as a plugin.
Here's where this gets bigger than one workflow.
You can package the whole thing as a plugin and hand it to your partners. They install it, and the plugin builds the folder structure, sets up the process, and wires up the skills for them automatically.
Now the plugin is the one thing that's unique to your org.
Everyone runs the same pipeline. Your process is standardized, and onboarding a new partner takes one install instead of a day of setup.
Think bigger.
This is a simple flow, but the concept underneath it isn't about due diligence at all:
Use a subfolder structure as the state machine for any repeated process. Feed it on a schedule. Run focused skills on each item. Chain them. Visualize the result.
All of it local to your computer, and the scale you can reach is unlimited.
Due diligence is my example. Yours might be hiring, content review, support tickets, lead qualification, contract review. The pattern doesn't change.
There are a million ways to use this. Start with the one process that eats your week.
Want to build your own? Paste this into Cowork, Claude Code, or Codex and it'll interview you, then build the plugin for you:
--
I want to build a plugin that automates a repeated process in my work using a subfolder pipeline. Here's the pattern I'm copying:
A main folder with subfolders for each stage an item can be in (e.g. Pending, Needs Review, Approved, Denied)
A scheduled task that feeds new items into the first stage automatically
A series of small, single-purpose skills that evaluate each item in the pipeline
One orchestrator skill that runs all the evaluation skills, decides the outcome, and moves each item into the right stage folder
An HTML dashboard that visualizes how many items are in each stage
The whole thing packaged as a plugin I can share with my team so they get the same setup with one install
Before you build anything, interview me one question at a time to figure out how this applies to MY work. Don't assume my use case is the same as the example. Ask me about:
What repeated process am I trying to automate, and what does "done" look like for one item?
What are the stages an item moves through? (These become my folders.)
Where do new items come from? (Email, a database, a form, manual drops?) How should they get fed into the pipeline, and how often?
What checks or evaluations should run on each item? Walk me through each one as a separate skill with its own pass/fail logic.
What's the decision rule for where an item ends up after evaluation?
What do I want the dashboard to show me at a glance?
Do I want to share this with a team, and should the plugin set everything up for them on install?
Ask follow-ups if my answers are vague. The goal is a plugin flexible enough to handle my process, not just the example above. When you have enough to work with, summarize the plan back to me, get my confirmation, then build it.
Start with question 1.
--
If you build something with this, reply and tell me what process you automated. I want to see what people put through it.
Bookmark this for when you're ready to build. Enjoy!
A US export control directive just pulled your model mid-sprint.
This is a contracts and continuity problem, not just a tech problem.
If your AI stack has one model and no fallback SLA, today is the day to fix that.
Opus 4.8 as a default. Local model as a floor. Multi-vendor as the architecture.
The benchmark leader has changed 3× this month.
That's not a sign to keep chasing the best model.
It's a sign to stop betting on one.
→ Build on the routing layer (MCP)
→ Add tools once
→ Swap models when they're ready
The abstraction layer outlives the arms race.
Anthropic's Claude Code creator says software engineering as we know it is ending.
It's true.
But here's the part people are skipping: the replacement is "builders" not fewer jobs.
→ Builders = domain expertise + AI orchestration
→ Not pure devs. Not pure lawyers. Not pure anything.
→ The hybrid knowledge worker is the job that survives.
I'm a lawyer who builds AI systems for his own firm and companies. I didn't plan to be early on this curve. But here's what I'd tell anyone watching: start shipping something with Claude Code this weekend. The learning compounds fast.
What are you building that you couldn't have shipped a year ago?
I shipped a custody tracking app and a golf side game app.