Anthropic and OpenAI battle for the future of AI coding
Plus: new MCP releases from Figma and Shopify, why product leadership is critical for responsible AI feature development, new tools and more...
This week’s briefing is supported by the new AI presentation tool Chronicle. It’s an impressive new way to create designer-grade presentations at work and it integrates with product team tools like Notion, Figma, Slack and others.
Hi product people 👋,
This week saw tech leaders from GitHub, Google, Anthropic and others all lay out their vision of the future - with some surprising insights from Google’s Head of Search on how AI is really impacting their core search business.
Plus, major new releases including GPT-5 and the battle of AI coding, an MCP product from Shopify that uses “MCP UI” and a new study which demonstrates the critical role of product leadership in developing AI responsibly .
Happy Friday and have a great weekend ahead!
Rich
Watch on YouTube | Follow me on Substack Notes
Anthropic gets sub-agents and battles with GPT-5 for coding benchmark supremacy
First up, Anthropic made some major announcements this week that could impact product teams. First is the release of Subagents in Claude Code. Subagents are pre-configured AI personalities that Claude Code can delegate tasks to. Each subagent has its own specific purpose and can be configured with specific tools that it's allowed to use. For example, a QA subagent could proactively investigate errors. Subagents can be created for any recurring or specialized task (e.g., test automation, documentation generation, compliance checks). Here’s a demo of it in action.
The second is an incremental upgrade to Claude Opus with 4.1. The upgraded model scores a 74.5% on the software engineering SWE test vs 72% for Opus 4. Yesterday, OpenAI unveiled its latest GPT-5 model and it scores 74.9% on the SWE test, up from o3’s 69.1%. GPT-5 comes with some impressive vibe coding abilities and OpenAI has published a gallery of apps you can explore. You can read more about the new model’s coding capabilities here.
Anthropic and OpenAI are essentially neck and neck with barely a decimal between them but according to new analysis, Anthropic is in a pretty risky position with almost~50% of Anthropic’s API revenue coming from just 2 customers: GitHub Copilot and Cursor. What happens now that GPT-5 beats Claude on the SWE test (albeit marginally)? Will developers using GitHub Copilot and Cursor switch to that? As developers get more time to experiment with GPT-5 we’ll start to get a better picture of what happens next.
Meanwhile…
Perplexity announced a partnership with OpenTable that lets users book restaurants from directly within Perplexity without ever leaving the app. It’s positioned as “powered by OpenTable” which is a win for them. But I do wonder how these types of partnerships will play out long term. Becoming the default partner in AI search products for specific niches could become pretty lucrative over time with the right revenue share model but as with all other parts of the web that are increasingly getting hoovered up by AI, it also risks destroying the direct relationship users have with brands. Ecommerce companies are starting to worry about this.
And speaking of ecommerce, Shopify is launching a series of new tools that use AI agents in its ecommerce stack. This includes a new checkout kit and the use of MCP UI - an extension of the MCP protocol which allows companies to embed images of their products directly inside AI conversational tools. In Shopify’s case, this means stores can embed images of their products inside AI tools and agents.
Figma also made some MCP updates of its own. The MCP server can now read annotations directly from your Figma designs. This means any notes about interactions, accessibility, or other design considerations are surfaced to AI agents when generating code. See it in action here.
Tools you can use
Chronicle - craft stunning decks in minutes. Chronicle pairs expert storytelling with AI to elevate your next presentation. It also integrates with tools like Notion, Slack, Figma and others - with professional looking results.*
North - the new AI agent platform from Cohere released this week. draft and refine documents, including PRDs, financial reports, market research, and sales pipeline reports, adhering to your style guide and formatting requirements.
Graphy - From messy data to beautiful graphs in a click, Graphy lets you quickly create graphs and customize them through a conversational interface.
Lindy - the simplest way for businesses to create, manage, and share agents. Lindy agents can now use their own browsers to complete tasks like collecting product feedback, user research and more.
*sponsored content
Key reads and resources for product teams
New from the Department of Product Substack this week:
Knowledge Series - How to develop an MCP product strategy
Just as they did with APIs, product teams will increasingly need to figure out what their own MCP strategy looks like. Not just for internal use - but for external clients, too.And that’s exactly what this Knowledge Series is designed to help product teams do. We’ll explore some of the latest MCP releases from companies including Notion, Stripe, Asana, Linear, Zapier and others to understand what capabilities these companies offer to end users. And then, based on some of these examples, we’ll consider what questions product execs and teams need to be asking themselves when crafting their own MCP product strategy.
(Department of Product)
Analysis - The future of software development
Vibe coding, prompt engineering & AI Assistants. Erik Torenberg interviews a16z’s Martin Casado, Jennifer Li, and Matt Bornstein breaking down how infrastructure is evolving in the age of AI from models and agents to developer tools and shifting user behavior. (A16z)
What tech leaders are saying this week
Google’s Head of Search says AI in search is driving more queries and higher quality clicks. People are searching more and asking new questions that are often longer and more complex. In addition, with AI Overviews people are seeing more links on the page than before, she says. (Google Blog)
GitHub’s CEO has warned that engineers should embrace AI or get out of engineering. It’s a pretty blunt assessment but is a sobering wake up call for engineers who still aren’t adopting AI in their daily workflows. Now, of course, if you want to play devil’s advocate here, GitHub has an AI coding product to sell so the more engineers who do adopt it, the better for them, but according to Dohmke, getting past the initial scepticism is critical for success. Engineers who do, ending up becoming more ambitious - and happier.
Cursor’s CEO told The Verge that he thinks 20–25% of a professional software engineer’s job could be fully delegated to AI, with the potential for this to rise to over 50% as technology advances. On vibe coding, he says that while some may use it to build their own software, most people will still use software built by a small, passionate minority (5% of the population).
Stripe’s co-founder spoke to Anthropic’s CEO to discuss AI business models, programming and product design. (Stripe YouTube)
📈 Product data and trends to stay informed
ChatGPT has hit 700 million weekly active users - up from 500 million weekly active users. This implies that OpenAI is now generating $1 billion a month ($12bn a year), compared to $500 million a month at the start of the year according to new analysis from The Information.
OpenAI is trying to manage its own success by rolling out new features that will prompt users to take a break from excessive ChatGPT use with gentle reminders.
Microsoft has published a major new study showing how people are using Copilot. The dataset analyzed includes 200,000 anonymized Copilot conversations from the U.S. over nine months in 2024. The most common use cases include summarizing information and writing. It also includes an “AI applicability” score - to estimate how relevant and useful generative AI is for different occupations. Some are using this as a proxy to figure out which jobs are most at risk from AI disruption. Read it here.
Reddit has reported 416 million weekly active users, up 22% year on year. Reddit Answers, the company’s new AI tool, reached 6 million weekly active users, up 5x on the previous quarter and its machine learning translation tool now supports 23 languages.
New analysis shows there is some cross-over in users of vibe coding apps. Nearly 21% of Bolt users also browsed Lovable over a three month time period. And, 15% of Base44 users also checked out Lovable. One VC says that churn for vibe coding products like Lovable is “huge”.
77% of product managers say they’re uncertain about what “responsibility” means when building new generative AI features or products. But product leadership heavily influences how PMs perceive responsibility. PMs were 2.3 times more likely to take actions such as testing for bias when working in companies where leadership teams showed a commitment to AI responsibility. Full study here.
Paid subscribers get the full DoP Substack including: The Knowledge Series for sharpening your tech skills, AI tutorials for putting AI into practice at work and DoP Deep dive reports to learn from the world’s top tech companies.
Not sure what those benchmarks are testing but GPT 5 is far behind Sonnet even! In coding at least.
https://open.substack.com/pub/growingupaspen/p/protecting-humanity-through-ai-ethics?r=2g93c&utm_medium=ios