Drate Berry

Building AI for growth marketing teams @OptimizePilot | @Harvard CS | Exploring research → product | Coffee & corgis

[GREG ISENBERG @gregisenberg]

The takeaway from Fable 5 being BANNED by the government: GET GOOD AT LOCAL MODELS SO YOU HAVE 100% CONTROL.

My entire weekend was going to be building my craziest ideas with Fable 5. That's now cancelled.

So instead of building with Fable this weekend, I've decided I'll go deep on local models:

1. Start with the runtime. Download Ollama or LM Studio first. This is the thing that actually runs models on your machine.

2. Match the model to your hardware. A model's size is measured in billions of parameters (7B, 32B, 70B). Bigger is smarter but needs more memory. Rule of thumb: a 7B model runs on almost any laptop, a 32B needs a good Mac with 32GB+ RAM, a 70B needs serious hardware like a DGX Spark or a maxed-out Mac Studio.

3. Know which model for which job. Qwen 3 is the best all-around choice for most tasks. DeepSeek for reasoning and coding. Gemma 4 when you need something tiny that runs on a phone. Llama when you want the biggest community and the most fine-tunes.

4. Quantization. You can shrink a model to run on weaker hardware with barely any quality loss. Look for versions labeled Q4 or Q5. This is how a model that "needs" a server runs on your laptop. Learning this one concept changes everything.

5. Connect it to your agent. Point Hermes or your agent stack at a local model.

6. Context window is your real constraint locally. Cloud models give you huge context for free. Local models make you pay for it in memory. A bigger context window eats RAM fast. Keep your sessions tight and your prompts lean or your machine chokes.

7. Learn to give local models tools. A smaller local model with web search, file access, and code execution beats a giant model with none. The capability gap closes fast when you wire up the right tools. The model is the engine but the tools are the wheels.

8. Fine-tuning is more accessible than you think. You don't need this on day one, but know it exists. You can take an open model and train it on your own data so it gets good at your specific domain.

I'll probably do a breakdown at some point on this @startupideaspod if people are into it.

The lesson from this ban is basically don't build your entire workflow on something that can disappear with a single letter. Own part of your stack. Local models are insurance.

It reminds me when people realized they don't own social media accounts. And then you saw people build email lists etc.

I remember running a startup and my biggest traffic source was organic FB. All of a sudden, algo changed, and I lost 99% of my traffic.

Same sorta moment (but bigger) for AI.

This is a wake up call.

[JH Scherck @JHTScherck]

Hate to be the guy coining a new "led-growth" term, but we're seeing a new type of growth across our customer base:

Vibe code-led growth.

When people build their own tools with Claude and Codex, they often need additional services. AI is telling people what to buy (and what to avoid!) when vibe coding. AI is also price sensitive and wants you to avoid sales calls and just get to building. Adjust accordingly.

[Anthropic @AnthropicAI]

The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.

The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.

Access to all other Claude models is not affected.

We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.

Read our full statement: https://t.co/bwn0sximKZ

[Drate Berry @drateberry]

So true. My rule for AI coding is that I should be able to do and understand everything I’m asking AI to do, and be able to understand the decisions it makes. If I can’t, then I can’t ship the code for even an internal project.

[andrei saioc @asaio87]

Nobody wants to admit it but "vibecoding" works until your app has actual users. Then you discover what senior engineers have been getting paid for.

[Drate Berry @drateberry]

@zdogmode Making an assumption this is a B2B startup. I think founding in yours 30s is perfect timing wise. You’ve spent time in industry getting to know how businesses work and you understand their actual pain points, can speak their language, and can leverage your network.

[Amazon Web Services @awscloud]

More AI-generated code doesn't make your team faster. It might actually slow you down.

[Drate Berry @drateberry]

Had to explain this to a prospect recently. They couldn’t understand why the AI solution they wanted would be 10x more than a Claude subscription.

I’ve started using the Uber model - most people remember when Uber was young and rides were super cheap, but as they pushed taxis off the streets in a lot of cities, their pricing went up. When VC funding slows (or dries up) the subsidized “free lunch” goes with it.

What looks cheap now and is causing companies to replace workers is going to end up being more expensive than the laid off workers once companies shift to a pure token based billing model.

I think we’ve had a good run, but the subsidized AI era is coming to an end as soon as the public markets begin to have influence on the companies’ roadmaps.

[SemiAnalysis @SemiAnalysis_]

Recently, we purchased one of each Anthropic/OpenAI subscription plan and randomly ran long horizon coding tasks until we exhausted the weekly limit. It's widely believed that a $200/month plan maxes out at ~$2000/month worth of tokens (assuming API pricing). However, we found that the subscriptions are actually far more generous. (2/4)

[Drate Berry @drateberry]

This has nothing to do with Shopify (assuming all of your evidence was submitted to the customer’s bank). Chargebacks get decided by the credit card issuer as they have to eat the cost if the merchant is unable to refund the money. That said, I pursued a small claims case against a customer in a similar situation. They paid with an Amex, which at that time was known for siding with the customer almost 100% of the time, but the product was delivered. They operated a fairly well known publication, had syndicated content deals, and book deals, so once I filed the small claims case, they sent a wire to not deal with it. Some people are just shitty and try to get free shit.

[Drate Berry @drateberry]

@asmartbear The only time I’ve ever claimed to be in stealth mode was when I was in the ideation phase. I know there are some great co’s that had a stealth phase, but the ones I’ve personally worked with that went that route are the biggest dumpster fires without any stable direction.

[Drate Berry @drateberry]

I get the concern @AnthropicAI has about using AI to exploit vulnerabilities, but it’s going to get really old if frontier models are hampered from running pen-tests on code that I own. Not sure what the solution is, but this is definitely a feature I’d pay a lot for and jump through a lot of hoops for access.

[Drate Berry @drateberry]

One of my biggest pet peeves is fast food places surcharging credit cards. As a business owner, I get that credit card fees exist, but it’s a part of doing business in the modern era. Also, if you’re a chain and paying 3% in fees for card present transactions, you are an idiot.

[Boris Cherny @bcherny]

Salesforce published a detailed writeup on going agentic with Claude Code. A couple things jumped out.

A migration they'd scoped at 231 days shipped in 13. One PR delivered 21 endpoints at 100% test coverage.

[Drate Berry @drateberry]

I watched Perplexity's Comet do data entry in Google Sheets this morning, and it's bad. So bad...

How are business leaders watching an agent fumble through a browser demo and concluding "I can replace my ops team" or "we don't need data entry clerks anymore"? They're not seeing the whole picture.

In a browser, every UI the agent navigates is built for ten fingers and two eyes. Dropdowns, modals, and pagination are scaffolding for humans. The agent has none of those limits, so it has to mimic them to use the browser: read the screen, find the cell, click, type, recheck, repeat.

Hundreds of tokens and several seconds per cell, for something an API could do in one request.

What appears to be a worker you can cut is actually an expensive workaround for infrastructure that doesn't exist yet. Klarna replaced 700 support agents with AI, then spent 2025 rehiring people after quality and satisfaction cratered.

So, before replacing humans, the bottleneck isn't headcount. It's that most current systems can't be operated by anything but a human.

Pointing agents at human interfaces isn't automation, and it isn't even cheap once you see the AI token bill.

The future isn't swapping humans for agents 1:1. It's rebuilding systems for efficient AI use on repetitive tasks, so humans can focus on higher-order work.

[Mark Ajzenstadt @mardehaym]

None of this is satire.

→ A company spent $500,000,000 on Claude in one month because nobody set usage limits

→ Uber ran leaderboards ranking engineers by how much AI they used, not what they shipped

→ Uber burned their entire 2026 budget by April. Their COO said he can’t connect any of it to consumer features

→ A CTO told Axios employees were using enterprise AI to check the weather

→ Microsoft canceled most Claude Code licenses because the token bill spiraled

→ Companies are now laying people off to pay the AI bill. Not because AI replaced the work. Because the bill replaced the headcount.

[Drate Berry @drateberry]

@toddsaunders Haven’t used it a ton today, but it does seem somewhat better at adhering to a given standard. There were times in 4.7 where I’d be reviewing code against a scope document and it would say everything was there when it wasn’t. So far, 4.8 called out a section I’d skipped over

[Drate Berry @drateberry]

Congrats to everyone at @AnthropicAI, but this is wild. Some of the most respected VCs in the world are betting that the AI market is going to keep growing. What a time to be involved in CS.

[Anthropic @AnthropicAI]

We've raised $65 billion in Series H funding at a $965 billion post-money valuation, led by @AltimeterCap, Dragoneer, @Greenoaks, and @sequoia.

This investment will help us advance our research and expand our capacity to meet growing demand for Claude.

[Drate Berry @drateberry]

Is it just me, or did @Mercury abruptly close a lot of accounts today? I just received notifications of account closure + checks issued for account balances within 15 minutes of each other.

[Drate Berry @drateberry]

Dear Boston, we really don't need to start the whole rain every single weekend thing again this summer.

[Drate Berry @drateberry]

@DagnyTaggartUSA @ijoshuajohnson Have Walmart deliver, just like you’re getting Whole Foods delivered. Also, Walmart is the largest organic grocer in the country, and that often comes with better pricing. Unfortunately, Whole Foods (and organic food in general) is expensive no matter what “color” your state is

[Drate Berry @drateberry]

I’ve been working on a blog post about the anti-data center movement I witnessed driving through Texas over Mother’s Day weekend. I’m now on my third outline, because there’s been a major new development of counties/cities blocking development everyday this week.