John McMahon

California just launched the Delete Request and Opt-out Platform (DROP) , giving residents a centralized way to submit a data deletion request and opt out of data brokers in one flow. Instead of contacting hundreds of vendors individually, a single DROP request now reaches 500+ registered data brokers under the California Delete Act and CCPA privacy rights enforcement. Why ? Data brokers collect and resell personal data, including identity signals, device identifiers, location history, and behavioral profiles. That data fuels spam, fraud risk, enrichment pipelines, and AI training exposure. DROP restores control back to consumers. Residents verify eligibility through secure identity gateways and submit identifiers such as name, ZIP code, email address, phone number, and optional device signals for better matching accuracy. Brokers must process deletion requests, honor opt-out requirements, and report status under California Privacy Protection Agency oversight. Analytics and growth teams should treat this as a signal to audit third-party data dependencies now. Any enrichment pipelines pulling broker data need clear legal basis alignment with CCPA and CPRA. Consent frameworks, identity resolution logic, and downstream activation models will need tightening as broker supply contracts. This also reshapes how first-party data strategy compounds over time. Clean consent. Transparent value exchange. Minimal reliance on opaque enrichment sources. Teams that are already invested in disciplined data governance will move faster than those still leaning on shadow data flows. Platform access: https://lnkd.in/dY6D7F6d How will this reshape your data privacy strategy and consent operations? #DataPrivacy #CCPA #DeleteAct #ConsentManagement #PrivacyCompliance #MarTech #DataGovernance #DigitalTrust

13

1 Comment

The Biggest Mistake Founders Make When Donating IP to Open Source (and how to turn a gift into a growth engine). When a scaling company decides to open-source its IP, whether it's an algorithm, a data standard, or an entire app, it feels like a pure philanthropic act. It isn't. It's a strategic weapon, but only if you ask the right questions. We often focus on the emotional "goodwill" of the donation, but the real value is in the non-monetary return and the legal clarity. If you’re considering giving away your IP, here is what you need to know before you push to GitHub: The Strategic Shift: From Cost to Contribution 1. Stop Thinking DONATION Start Thinking STANDARD: You're not just giving away code; you're attempting to set the de facto industry standard. If the World Health Organization (WHO) or a key industry body adopts your code, you gain immense credibility and influence that money can't buy. • The Return: You get to shape the narrative and the technical roadmap for an entire sector. 2. The "Tax" of Exclusivity: Global bodies like the WHO or World Bank will not accept an exclusive license. They need the freedom to share the tech globally. Your commitment must be to a Non-Exclusive, Royalty-Free license (i.e., true Open Source). Don’t waste time trying to negotiate control you don't need. 3. The Competitor Clause: Open Source means your competitors can use your code. That's the point. If your business model relies on secrecy, don't open-source your core IP. If your model relies on service, integration, and community around the IP, then Open Source is your cheapest marketing strategy. The Legal and Technical Must-Haves 1. Choose Your Legal Armor Wisely: Not all Open Source licenses are equal. • The MIT License is the simplest. Perfect for rapid adoption. • The Apache License 2.0 is comprehensive. It includes an explicit Patent Grant, which provides vital legal protection for the users of your code (governments, large companies) against future patent claims. For any serious B2B/global contribution, Apache 2.0 is often the safer choice. 2. The Maintenance Burden: Who owns the bug fixes? When donating to an external body, you must have a clear, written agreement on who will handle long-term maintenance, security patches, and support. Otherwise, your donated app becomes a liability to the recipient organization. 3. Clean Your Commit History: Ensure the code base you release does not contain proprietary secrets or accidentally licensed third-party code. Legal review is mandatory. A "dirty" IP donation can boomerang into a massive liability for your company later. Ask yourself: Is your goal to get a nice press release, or to have your IP become the foundation that the next generation of your industry builds upon? I'm actually working on IP donations to some global organizations and I thought I should share this info. #OpenSource #IntellectualProperty #DigitalStrategy #ScaleUp #FounderTips #TechForGood

5

In terms of LLM use cases, one of the highest value add uses is software coding, and in this application, the proprietary models dominate: "proprietary models still handle the bulk of coding assistance overall (the gray region), reflecting strong offerings like Anthropic’s Claude." This is especially important because coding is now the dominant, token-weighted use case — in early 2025, coding represented 11% of token volume, but toward the end of 2025, this share exceeded 50%. The following exhibit from the report highlights proprietary model dominance in coding use cases:

1

Reading Metronome’s Monetization Operating Model, I kept coming back to one idea: pricing has become product. Software now delivers outcomes, not access. Yet most companies still charge as if they’re selling seats or licenses. That disconnect creates friction: for customers, unpredictability; for companies, stalled growth. The paper’s argument is simple but sharp—monetization isn’t a late-stage decision. It’s strategic infrastructure. Pricing needs the same ownership and iteration as any feature. Treat it like a surface that customers touch, not a spreadsheet buried in finance. If value is continuous and dynamic, pricing must be as well. That means product, GTM, finance, and engineering working from one system of truth. How many of us still treat pricing as an afterthought—when it should be a growth engine? https://lnkd.in/gnH7WzYf #Monetization #ProductStrategy #AI

8

There is an interesting question raised by Vincent Laufer, MD, PhD's post - can a "black box" prediction made by AI be useful, if it doesn't disentangle mechanisms? Let's explore this in the context of cancer patient treatment and how we predict response to therapies... Drugs approved typically are based on biology, and the flow is usually from proposed biological relevance to MOA to treatment effect. However, there are cases when drugs were approved when completely new indications were found during clinical trials. Sildenafil (Viagra) and minoxidil (Rogaine) are two of the most famous examples. In this case, at DatAnon we are exploring more on the side of drug repurposing, or more precisely determining any subgroups of patients which are predicted to have better average response for a given therapy. It is possible to look at this from the other side, ie. which drugs are predicted to be most efficacious for a given patient and can be then used in functional diagnostics panels to test on the patient's own tumor. That is the goal of the first sets of analyses that we will be launching, which will examine huge numbers of compounds in the SOC, oncology and non-oncology compounds, and even OTC/supplements based on the gene-expression signature of the patient's tumor. Can we disentangle predicted subgroups of response to MOA and biomarkers? That requires a bit of detailed explanation, so I'll save it for other posts!

5

1 Comment

🐮 AI-powered cows worth $2 billion are using an algorithm called the "cowgorithm." I'm not joking. Halter makes AI-powered cow collars. They monitor health, location — and herd cattle. Farmers tap one button on an app and cows gather for milking. 600,000 collars already live. Peter Thiel just backed their latest round at a $2 billion valuation. Every time I think I've seen the most ridiculous application of AI — I'm proven wrong. But here's what the headline misses. This isn't a funny story about cows. It's the proof of concept for the entire physical world AI thesis. The collar is solar-powered. GPS. Health monitoring. Heat cycle detection. 6,000+ data points per minute. Per cow. That stream feeds the "cowgorithms" — models that translate farmer intent into signals the animal understands, and animal behavior into decisions the farmer can act on. No ranch hands. No horses. No physical fences. One app. One button. 600,000 cows moving. The number that reframes everything: Traditional fencing: ~$20,000 per mile. Halter customers have created 11,000 miles of virtual fencing in the US alone. $220 million in avoided costs. From collars. The valuation trajectory is the real signal: 2023 → $85M. Proof of concept. June 2025 → $1 billion. 300,000 collars. March 2026 → $2 billion. Founders Fund. 600,000 collars. Valuation doubled in 9 months. Collar count doubled in the same window. Thiel backed Palantir when defense software was unglamorous. SpaceX when rockets seemed too physical. Now Halter — when AI collars for cows seem too ridiculous. The $7 trillion global agriculture industry is the largest physical-world industry on the planet that has received the least AI investment relative to its economic weight. Halter isn't a cow collar company. It's the first proof that the physical AI stack — understand current state, predict future state, control future state — works on biological systems at industrial scale. The cow is the sensor. The collar is the compute. The cowgorithm is the model. If 600,000 collars can autonomously herd cattle — what does 6,000,000 look like? What does that dataset become when it's large enough to predict disease outbreaks, optimize global protein supply chains, and inform regenerative land management at continental scale? The cowgorithm today is a herding tool. The cowgorithm at scale is a biological world model. The physical world is 100% of the world. We've barely started digitizing it. What's the "cowgorithm" in your industry — the problem that seems too ridiculous to take seriously, but is actually the $2 billion opportunity nobody is looking at yet? https://lnkd.in/eiHTsrcC 🐮

5

Running a Company on Markdown Files CompanyOS: a skills-only system that turns Claude Code into the operating layer for an entire company. No application code, no web UI - just markdown files that teach Claude Code how to run business operations. https://lnkd.in/gACFhTMk #AdventuresInClaude #ClaudeCode #CompanyOS

268

52 Comments

Most vibe coders are optimizing the wrong layer. There’s a post going around showing how a “non-technical” prompt and a “senior engineer” prompt produce wildly different results for the same task. The takeaway people grabbed: write longer, more detailed prompts. That’s not the lesson. Here’s what usually happens. A deal team needs to review 6,000 documents in a virtual data room for a mid-market acquisition. Someone asks an AI to flag every contract with a change-of-control provision. They get a script that searches for the phrase, pulls matches, and dumps them into a spreadsheet. Half the results are boilerplate. Key risks in assignment clauses and consent requirements get missed entirely. Nobody knows what the AI skipped or why. An experienced builder doesn’t write a better search query. They ask different questions first. What document types are actually in this data room? Are we looking for explicit change-of-control language, or any provision that could block or complicate closing? Do amendments override the base agreement? Are side letters in the set? What’s the priority: speed to first-pass flags, or completeness for the disclosure schedule? What happens when a clause is ambiguous and needs deal-team judgment? That reframes the problem before a single line of code gets written. The gap between a weak AI output and strong output isn’t verbosity or technical vocabulary. It’s the quality of the constraints. Success criteria. Tradeoff decisions. Knowing what not to build. Vibe coding gets dramatically better when you stop describing the task and start defining the boundaries around it. I Try this: next time you prompt an AI to work through a diligence set, write three sentences before the request. What’s a real risk versus noise? What should get escalated instead of auto-flagged? What’s the consequence of a false negative? That alone will change the output more than any prompt template. Subscribe to PossibLaw. Substack. Podcast. Tools to become a builder. We run custom training for legal teams who are done watching from the sidelines. Let’s talk.

1

2 Comments

We hear a lot about open source vs. closed source… Has OpenClaw settled this debate, or is it all hype? It’s clear that larger LLMs missed out on this one, so where do they go from here?

60

17 Comments

https://lnkd.in/gUK5W5YT  Truly open source AI isn't quite where it needs to be. Couple that with the expense of GPUs/hardware and a true "democratized" and private LLM will be difficult. As in all new tech, private AI will start with the enterprise and work its way down.

4

Worth paying attention to... Anthropic vs. Cursor. While Cursor was paying Anthropic hundreds of millions in API fees to power its AI code editor, Anthropic seems to have studied exactly what developers wanted, then launched Claude Code to "eat" their biggest distribution partner. The result. Claude Code hit $1B ARR in just six months and has already surged past $2.5B, proving that when you're a "wrapper" startup, you're often just paying your provider to learn how to replace you. Keep an eye on Cursor layoffs...

88

13 Comments

Sustained ~252–291 GiB/s of HBM memory throughput on H100s under decode load — essentially hitting the hardware roofline. This matters because HBM throughput, not FLOPs, is the real bottleneck in LLM inference. By keeping memory nearly fully saturated, I’ve unlocked far higher efficiency and throughput than standard engines. The result: 0.9–1.36M tokens/sec with ~0.1 ms first-token latency  #AI #LLM #GPURouter #H100 #Inference #CostEfficiency #Innovation #DeepLearning #AIInfrastructure #HighPerformanceComputing NVIDIA AMD OpenAI Google Shilpa Kolhatkar Keith Strier a16z speedrun

5

It's crazy how quickly things evolve in AI world. A few weeks ago I was setup OpenClaw, then Nanoclaw (ht: 👾 Andrew Draper) and now Claude has essentially replicated everything natively. With Claude Channels (released yesterday) you can now setup a private messaging channel with your agents - they can push messages to you and you can chat with them like a personal assistant. Here's how you set it up: 1. Install Claude - ideally on a remote server. I have it on a DigitalOcean droplet with 2 GB of memory - only $12 / month. You can use whatever you like - there are less expensive options available. 2. Install my open source agent framework from https://lnkd.in/eSBRDwh2. Don't know Git? That's what Claude is for - just ask Claude to install it. 3. Setup a Claude channel following instructions at https://lnkd.in/e7uBZ_YA. Don't know how to do that? Just ask Claude to do it! 4. You can now send messages to your main agent, which can dispatch things to more customized agents. It's not hard to do - I guarantee you'll be able to get this done. Chatbots were helpful - having a personal assistant one message away to help you get things done is taking AI to the next level.

26

12 Comments

I love Rails. I've built my wealth on it. That's exactly why I think we need to have an honest conversation about its biggest blind spot. Eileen M. U. gave a brilliant talk at Rails World last year: "The Myth of the Modular Monolith." After 6 years of modularizing Shopify's monolith, she concluded: "You can't solve human and cultural problems by changing your architecture." She's right kinda of about the culture part. But stopped one step short. Rails doesn't just allow tight coupling. It incentivizes it. ActiveRecord makes every model globally accessible from anywhere. Any controller, any model, any job can reach into any table through any association chain. There's no boundary enforcement. No import restriction. No compilation step that says "this module shouldn't know about that one." When your framework makes coupling the path of least resistance, blaming developers for coupled code is like blaming water for flowing downhill. Shopify tried to fix this with Packwerk. It just give you a map not a real solution. Rails Engines? Been there, done that. They are miniature apps that provide functionality to their host applications. They don't enforce real isolation and they weren't designed for progressive extraction. But this isn't a Rails problem. It's a missing tool problem. Look at what Nx did for the Node/TypeScript ecosystem: → Module boundaries enforced at the import level, not via YAML to-do lists → Dependency graph visualization built in → Affected-only CI: only test what actually changed → Clear public API per module, everything else is private by default → A real path from monorepo module to extracted service Nx didn't replace Node. It gave Node developers the guardrails to build modular systems naturally. The right thing became the easy thing. By default. Rails deserves the same. Imagine a Rails-native tool that: • Enforces module boundaries through autoload paths, not opt-in linting • Gives each module a public API (explicit exports) while keeping internals private • Generates a live dependency graph from actual code usage, not manual YAML declarations • Runs affected-only CI based on that graph Not a replacement for Rails. A companion that makes Rails' conventions work at scale. Eileen is right: culture matters. Education matters. Leadership incentives matter. But culture alone didn't give us CI/CD. Tools did. Culture alone didn't give us code formatting standards. Linters did. Culture alone won't give us modular Rails applications. Rails gave us developer happiness for building products. Now we need the same philosophy applied to building organizations of code. The myth isn't the modular monolith. The myth is that we have the right tools for one. Who's building this? 🎥 Eileen Uchitelle — "The Myth of the Modular Monolith" (Rails World 2024) https://lnkd.in/dabvzHQm

13

🤖 Anthropic Launches Claude Opus 4.6: Beyond Just Coding Anthropic just released their smartest model yet, and this time they're targeting more than developers. Claude Opus 4.6 is designed to handle complex, multi-step tasks with production-ready quality right out of the gate. Key upgrades include: - 1 million token context window (beta) - Enhanced capabilities in PowerPoint and Excel - Stronger performance in financial analysis and search - "Agent teams" feature for coordinating multiple AI agents on one project The focus? Expanding beyond coding into broader knowledge work - research, marketing, presentations, and more. For developers, the model promises to compress days of work into hours through Claude Code, handling everything from architecture to deployment. Pricing stays the same as the previous version, available starting today. Anthropic also ran their most comprehensive safety testing to date, including new evaluations for cybersecurity threats and harmful action detection. Are you using Claude for non-coding tasks yet? What's been your experience?

5

2 Comments

Dan Shipper and the team at Every Inc. coined a term that perfectly describes what we've been building at case.dev: compounding engineering. The idea is simple but powerful: in traditional software development, each feature makes the next one harder to build. More code, more complexity, more edge cases. But with AI-native workflows, you can flip this equation—each feature makes the next one easier. Here's what this looks like in practice for us: Our case.dev website is built directly from our source code repository. When I want to add a new API endpoint that we've deployed, I have a custom Claude skill that handles the cascade of work that used to eat up days: → Updates the actual case.dev website → Updates the documentation → Drafts the blog post announcing the feature → Updates SEO metadata → Runs sanity checks across docs, SDK, and code examples The learnings from each deployment get codified back into the system. The agent gets smarter. The process gets faster. The result? When a customer asks for a new feature, we ship it in hours, not weeks or months. This is what building developer infrastructure in 2026 looks like. The bottleneck isn't coding anymore—it's deciding what to build next.

8

Breaking: Claude Code just introduced Sub-Agents You can run your own custom Agents. Using the /agent command Based on an agent file format, written in markdown. Custom system prompts, custom tools. Making it very easy to create your own agents. Previously I made a REST api for Claude Code, meaning so I could script things from a browser. Now, I can create teams of sub agents, working together on a problem. More info on the feature: https://lnkd.in/g3H4_q7E You would think they are paying me, I'm just hyped because this allows Spec Driven Development to be even easier.

73

23 Comments