Robert Sweeney

Today we announced an upcoming change to how Augment Code is priced. Starting October 20, 2025, we’ll move from plans based on user messages per month to a credit-based model, where every plan includes a pool of credits that can be spent across different activities. Our current model treats every interaction as equal — but they’re not. A quick CSS fix isn’t the same as refactoring a mobile payments flow and adding Apple Pay. These tasks differ dramatically in scope, reasoning, and compute cost, yet today they’re priced the same. That’s not transparent for developers, and it doesn’t reflect how modern reasoning models behave. The best models pause to ask clarifying questions to produce better results. Under a message-based plan, that quality actually costs users more. It’s also unsustainable on our side; going forward, we're aligning pricing with real usage to maintain long-term reliability. A credit model fixes both problems: + Developers pay for actual compute and complexity, not arbitrary message counts + Augment stays sustainable while supporting enterprise-grade workloads and investing in the best Context Engine in the world For some customers, this will mean higher costs. Over the coming weeks, we’ll share tools to help teams forecast usage and understand what this means for them. Our goal is a fair, transparent model that supports what Augment does best — helping developers build with AI at scale. Read more on our blog → https://lnkd.in/gZdPTGjG

172

9 Comments

Today, we’re thrilled to publicly launch Phoebe and announce $17M in seed funding, in rounds led by GV and Cherry Ventures. We are building the immune system for software: An AI agent platform that continuously investigates live data, diagnoses emerging issues and generates preemptive fixes. As AI accelerates how software is built, reliability must keep pace. Phoebe started with faster troubleshooting, and our aim is to prevent the overwhelming majority of emergent problems before becoming customer-impacting failures. Huge thanks to the hundreds of early users at customers such as Trainline and PPRO. Thanks to our investors, including 20 CTOs and tech veterans joining as angels. And of course, thanks to the small and mighty Phoebe team! This is just the beginning! 🚀 Shoutout to Mark Kleinman at Sky News for breaking the news this morning (linked in the comments) Some of the many backers we're grateful for 🤗 🙏 GV (Google Ventures), Cherry Ventures, Anthony Danon, Carmen Alfonso Rico 🍫, William Gaybrick, Shola Akinlade, Yaron Shaer, Christian Dorffer, Tom Stafford, Thomas Wolf, Andrew Tan, Victor Riparbelli, Harsh Sinha, Jeanne DeWitt Grosser and more

681

139 Comments

A nice summary here from Messari on what we're up to at VeChain. The TL;DR is: We're showing up other L1s when it comes to real, mass adoption. In Q2, two of our VeBetter applications, GreenCart and Mugshot, each surpassed one million users, offering real proof that web3 can support everyday sustainable behavior at scale. The Stella Pay Visa Card now allows users to spend their VeBetter rewards at more than 130 million vendors around the world. You do something good, you earn something real, and now you can use it all over the world. This is what Web3 should be about. Bringing positive change to millions of people. No crypto native knowledge required. Messari's full report: https://lnkd.in/gTmkg7dw 

75

6 Comments

Prompts now sit at the heart of GenAI systems, moving through a flow that demands real engineering control. The process begins with creating and structuring prompts, adding parameters, configurations, and metadata. It continues with version control, where changes are tracked and teams collaborate. Automated evaluations, metrics, and quality gates then validate performance and reliability. Finally, prompts are integrated into production pipelines, connected to CI/CD, experiments, and monitoring. This flow is not static. Each stage feeds back into the next, enabling constant refinement and higher confidence in the system. Prompt management at this level requires engineering rigor, not one-off creativity. It is the foundation for scaling GenAI safely and effectively. What steps have you taken to make prompt management part of your engineering practice?

93

4 Comments

Here's how Lyft scaled from ML to GenAI Infrastructure without reinventing the wheel. Full video: https://lnkd.in/diNgXV6B In a standout talk, Konstantin Gizdarski gave a rare behind-the-scenes look at how Lyft expanded its mature ML platform to support LLMs and why that architectural choice mattered. 🎯 Key Insight: The breakthrough wasn’t starting from scratch; it was treating LLMs as a special case of ML models, and evolving the infrastructure accordingly. Here’s what stood out: 🔌 Unifying LLMs with Existing Serving Systems Rather than spinning up a separate stack, Lyft wrapped vendor APIs like OpenAI into their internal ML serving platform using custom proxies and client wrappers. This allowed them to: - Centralize observability, compute, and security - Remove client-side API keys - Gain full traffic attribution (who’s prompting what, where, and how) 🧪 Built-In Evaluation as a First-Class Citizen With LLM usage scaling rapidly, Lyft built its lightweight eval framework for: - Online input filtering (e.g. PII redaction) - Online output moderation (e.g. content safety) - Offline request–response scoring (LLMs evaluating LLMs) 🤖 Building Toward Assistants To support complex workflows, they’re developing an "Assistant" abstraction declarative interfaces that combine: - Core LLM functionality - Evaluations + proxies - Knowledge bases + tool use 🧠 Lesson: ML features → AI knowledge bases ML monitoring → AI evaluations By recognizing this conceptual symmetry, Lyft extended its platform rather than replacing it. 📌 Takeaway: Real GenAI adoption isn’t about picking the “right” model; it’s about platform architecture, user control, and infrastructure that scales with the org.

76

4 Comments

Today in “Scott Learns How Silicon Valley Works”: I’ve apparently been quoting ARR wrong this whole time. At Metronome we run a usage-based business. Customers sign up for committed contracts, find value, expand, and then negotiate new ones. It’s a virtuous cycle of NRR growth. To stay disciplined, we usually talk about cARR - the literal book value of our signed contracts. It’s conservative by design. But here’s the truth: most customers expand well beyond that number, often for multiple years. The confusion: I recently told an investor, “Our cARR is $X.” They assumed that meant a run-rate from last month’s revenue - and mentally discounted it by 50%. In reality, history says that $X in cARR is more likely to yield ~150% of X in actual revenue. A 3x swing, all in the framing. That’s just part of the fun (and pain) of building usage-based businesses: explaining why NRR, not ARR, is the real determining force. If you’re wrestling with usage-based monetization, let us know - we’ve had to learn this the hard way.

51

12 Comments

Two long years ago, at Movable Ink's Think Summit 2023, I shared how software developers have gained sizable leverage writing code over the past few decades. While there are only 1.6 million developers in the US, there are 160 million workers who haven't witnessed the same advances. But that was about to change! Fast forward to today. We're seeing AI transform companies in three fundamental ways: ⏰ We're nearing the end of humans writing software. Engineers, product managers, and designers are prompting AI to generate code, translate customer interviews into PRDs, and rapidly prototype new designs. 🛠️ We're capturing multi-step, manual, business processes and turning them into agentic workflows. Sales outreach copywriting, data transformation and sharing, prospect research, slide creation, and much more will be things of the past. 💻  And finally, the software that we build will be AI-first. Traditional UIs and user flows can break out of the constraints of human interaction. Think digital agents completing tasks or predictive recommendations eliminating entire steps. An individual contributor role will look more like an agent manager and those who learn how to harness these new tools will 10x their productivity relative to peers who don’t. Every business worker is about to experience leverage first-hand in a way that only developers have and skillfully managing agents will eclipse the ability to manage people. It's both a scary and thrilling time to be alive! 

42

1 Comment

Elie Schoppik has spent his career at the intersection of education and engineering. He built early ed-tech tools that helped people actually retain knowledge, founded a Bay Area coding bootcamp that trained thousands of developers through practice and feedback, and now leads Technical Education at Anthropic. Despite all that experience, Elie says most teams are using AI wrong: they’re treating it like an all-knowing oracle instead of a tutor. This is why Elie designs prompts, examples, and workflows that treat AI as a teacher, instead of a shortcut. In our conversation, we discuss: • Why relying on AI for “fast answers” erodes judgment and verification • How to turn prompts into guided lessons • Why context engineering is the most underrated skill in 2025 • Practical exercises any team can run tomorrow to raise AI fluency • What leaders should steal from Anthropic’s approach to technical education  Episode is live now — full conversation linked in the comments.

30

2 Comments