🔵 Claude Plugins and the SaaSpocalypse: is this “the most illogical thing in the world?”

Plus: Perplexity plots a comeback with impressive new UX research tools, worrying new data for ChatGPT mobile and how Meta and Google are driving internal AI adoption

Hi product people 👋,

Welcome back - and happy vibe coding anniversary. This week marks the first year anniversary from when OpenAI’s co-founder first coined the phrase, and one year on, the threat of vibe-coded and AI-powered products has sent global markets into a tailspin.

Nvidia’s CEO says it’s “the most illogical thing in the world” but after the release of Claude’s plugins, SaaS and other sector stocks plummeted this week. We’ll take a look at what it all means for product builders - and how you can actually use one of those plugins for product work.

Plus: Perplexity plots a comeback with some new features that could improve your UX research and design decisions, new mobile data makes for worrying reading for OpenAI and Anthropic unveils its vision of a new agent-powered development process.

As always, let me know if you have any thoughts in the comments below!

Rich

Watch on YouTube | Follow on Substack

---

Key reads and resources for product teams

New from the Department of Product Substack this week:

Knowledge Series - How to use Claude Code to create a board room of world-class product mentors

Build clones of Steve Jobs, Demis Hassabis and more to sit on your product board of mentors. It’s slightly dystopian, but Claude Code is now increasingly being adopted by non‑engineers, and its ability to reference markdown persona files, as well as capabilities like Agents and Skills, makes this idea possible in 2026 for product teams who want to occasionally lean on some world‑class expertise for decision making - just out of curiosity, if nothing else.

New in the AI library - Market sizing opportunity analysis

This prompt is a structured framework for conducting market sizing analysis for a new product or service. It guides you to systematically calculate three levels of market opportunity: TAM (Total Addressable Market), SAM (the serviceable portion for a specific region/segment), and SOM (Serviceable Obtainable Market - realistic first-year capture). It uses data sources like industry reports, competitor financials, and search trends to ensure your research is grounded in real-world data. (Department of Product)

Strategy - How to use generative AI for pricing

MIT says LLMs can make sophisticated pricing recommendations, but effective use requires users to provide well-crafted prompts and understand the tools’ limitations. (MIT Sloan)

Design - How to build a design system with Figma MCP and Cursor

Monday’s product design lead Elad Mizrahi explains how he uses AI to build and maintain a robust design system with far less manual handoff work. He replaces long, annotated Figma specs with concise documents written for AI, then uses Figma MCP to analyze designs and generate code-ready specifications. (YouTube)

Tools you can use - Supaboard

Connect to 600+ data sources, ask questions in plain English, and instantly generate dashboards and reports your whole team can trust.

Case study - How one tech company built 5 apps with Lovable to reduce costs by 99%

I have no idea if this person is affiliated with Lovable so I take this with a pinch of salt but this fascinating case study explains how one tech founder replaced his entire SaaS product stack with Lovable across different business areas including: sales, project management and call scheduling. Building these apps is easy today but the important deciding factor here will be how easy it is to maintain these apps once they’re built. (X)

How Meta and Google are driving internal adoption of AI

Meta is formally baking AI usage into how employees are evaluated, rewarded, and organized. At a recent companywide meeting, Meta reportedly told staff that performance reviews and bonuses are now closely tied to how effectively they use AI. The centerpiece is an internal, AI-powered performance tracker called Checkpoint, which pulls data from tools like Google Workspace and engineering systems. For engineers, it looks at AI-generated lines of code, error and bug rates, and how much code is written with vs. without AI, along with 200+ other metrics. Alongside this, Meta revamped its bonus structure to heavily reward top “AI-driven impact.” Employees in the highest of four performance tiers can get a 200% bonus multiplier, and a new “Meta Award” offers a 300% multiplier for standout performers.

Google, meanwhile, is running an internal initiative called Project EAT to drive AI adoption internally. It started as a grassroots effort in May 2025, run via the AI2 infrastructure org, and focuses on dogfooding Google’s latest AI tools (especially coding and productivity). A 12‑week pilot reportedly improved developer velocity, reduced workload, and enhanced code quality, and infrastructure-led adoption is seen as strategically critical because “infrastructure controls the pipes”.

---

New product features and innovation this week

SaaS stocks have taken a further battering this week amid fears that AI will replace SaaS products and right now there seems to be two camps emerging: investors who think the replacement of enterprise SaaS products is inevitable and those who think the idea is absurd. This week, Nvidia’s CEO said it’s “the most illogical thing in the world”.

But what’s actually happening in the real world?

On the consumer side, the idea that the average consumer will want to build and maintain their own apps has always seemed pretty far-fetched (and highly unlikely), but there are some early signals that consumers may find the idea of building mini apps appealing. A new social network, that’s been described as TikTok for mini apps, has started to gain significant traction. Gizmo is a social network feed of small, interactive mini apps that users can either use or build themselves - all through prompting. It’s been downloaded 600,000 times so far with around 250,000 of those coming in December alone. And with over 12,000 ratings it’s currently at a very impressive 4.9 stars.

Ramp’s product manager says the company has quietly built their own revenue tools internally to help their sales teams drive more sales. It’s powered by their own customer data platform together with a series of AI agents embedded directly into their workflows. It’s another warning sign for SaaS companies that AI-native teams are more than capable of building the exact tools they need at record speed. The idea that the average consumer will want to build and maintain their own complex apps seems far-fetched (and highly unlikely), but for enterprise, if AI-native companies like Ramp can very quickly build and maintain their own internal products that drive important strategic results (like driving revenues) then surely the argument that building internal products isn’t worth the hassle doesn’t stand up to scrutiny for long?

Perplexity’s comeback? A new Deep Research tool can help with UX research queries and Model Council

Perplexity has had a pretty rough few months as it struggles to differentiate itself but this week, it released a major new Deep Research feature that it says beats all others across dimensions including UX design. As part of the release, they published a new benchmark they call DRACO (Deep Research Accuracy, Completeness, and Objectivity (DRACO) Benchmark). DRACO is a curated set of 100 difficult research tasks that come from real user queries to Perplexity’s Deep Research product. These tasks span 10 domains (like finance, medicine, law, technology, academic research, product comparison, UX, etc.) and intentionally pull in information from sources in 40 different countries.

The UX tasks include realistic design questions a product team might ask, such as researching how the timing of AI code suggestions in the UI affects developer flow, or synthesizing evidence about UX patterns and user behavior. For each UX task, domain experts create a detailed rubric: they spell out what facts need to be correct (for example, what various tools actually do), what kinds of comparisons or trade‑offs a good answer should include, how clearly the answer should be structured, and how well it should cite sources like productivity studies or UX research. Deep‑research is then run on those UX tasks, and an LLM judge checks each answer against the rubric.

For product teams, the new research tool could be helpful for validating design decisions before building.

The company also unveiled what it calls “Model Council” which is a new feature that compares the output of multiple frontier models at once and identifies gaps between them so that users can decide which one to use. Strategically, this is a neat feature from Perplexity, since it leverages one of their only differentiators: the ability to use and compare multiple different models. But as models increasingly converge, it’s a moat that may not be sustainable.

Anthropic’s Cowork gets stronger with Google integrations and plugins; Claude’s PM says Cowork is transformative for managing user feedback

Anthropic has added some new integrations to Claude. Cowork now supports Google Drive integrations which means you can use it to work directly on Google apps like Gmail, Calendar and Docs. Plus, you can now connect your Slack channels to Claude to draft messages, send messages and create canvases that can be shared in specific channels.

Claude Code’s product manager Cat Wu says that the introduction of Claude Code in Slack has changed how quickly they respond to user feedback and ship product improvements. Anthropic has a user feedback channel where they regularly tag in Claude to investigate issues and push fixes.

Claude has also released “plugins”. Plugins are ready-made bundles that let you customize how Claude works for your role, team, and company when using Cowork. Each plugin combines skills, connectors, slash commands, and sub-agents into a single package.

How product teams can use Claude Cowork plugins

As part of the release of Claude plugins, Anthropic has published a free product management plugin . The updated version of Claude Cowork lets you browse and install the plugin directly in the Mac app so you can install it directly from the Cowork interface if preferred:

Here’s some of the ways you can use it:

• Feature specs & PRDs - generate structured product requirements documents from a problem statement or feature idea. Includes user stories, requirements prioritization, success metrics, and scope management.

• Roadmap planning - create, update, and reprioritize your product roadmap. Supports Now/Next/Later, quarterly themes, and OKR-aligned formats with dependency mapping.

• Stakeholder updates - generate status updates tailored to your audience (executives, engineering, customers). Pulls context from connected tools to save you the weekly update grind.

• User research synthesis - turn interview notes, survey data, and support tickets into structured insights. Identifies themes, builds personas, and surfaces opportunity areas with supporting evidence.

• Competitive analysis - research competitors and generate briefs with feature comparisons, positioning analysis, and strategic implications.

---

📈 Product data and trends to stay informed

Microsoft has revealed it now has 15 million paying users of 365 Copilot. At $30 per person, Microsoft is likely on track to generate several billions of dollars annually from these add-on subscriptions. Overall, this is a 3.3% conversion rate but investors weren’t particularly happy about that vs the amount of money the company has invested into Copilot.

According to the Wall Street Journal, from last July through late January, the percentage of Copilot subscribers who use the product as a primary option decreased from 18.8% to 11.5% which is disappointing but still, Microsoft raking in billions from its AI cloud infrastructure products, so consumer traction is arguably less important.

Grok MAUs web visits are up 29% on the month and ChatGPT has returned to growth after two consecutive months of decline. But, the ChatGPT mobile app is struggling. New data shows ChatGPT’s US App market share fell from 69.1% to 45.3% between Jan. 2025 and Jan. 2026; Gemini rose from 14.7% to 25.1%, Grok rose from 1.6% to 15.2%.

Claude Code installs by developers have exploded since December, reaching 35,000 daily installs vs ~20,000 for Codex and ~10,000 for Google Gemini. 4% of all commits on GitHub are now using Claude Code.

MIT researchers found that “real-time businesses” – organizations that can make decisions and act immediately on current data – achieved 20.6% higher revenue growth and 18.8% higher profit margins. Case studies from the research include United Airlines, which now automatically holds connecting flights using AI; IKEA, whose real-time algorithms cut delivery times from four weeks to two days; and Vanguard, whose proactive nudges led 100,000 investors to move $6.2 billion out of cash. For product teams, real-time capabilities are proven revenue drivers worth exploring.

Anthropic has published a major new report called the Agentic Coding Trends report which looks at how the software and product development process is evolving in 2026:

You can read the full report here but here’s some predictions and tidbits that might be of interest for product teams:

• Despite the hype, developers report being able to “fully delegate” only 0–20% of tasks, even though AI is used in roughly 60% of their work

• Roughly 27% of AI-assisted work consists of tasks that wouldn’t have been done at all - such as fixing minor “papercut” bugs or building nice-to-have interactive dashboards.

• AI primarily increases output volume (more features, more experiments) rather than just making existing tasks faster.

• Engineers evolving from “implementers” to “orchestrators” who focus on architecture, strategy, and system design while AI handles tactical implementation

• Technical debt that accumulated for years will get systematically eliminated

• Non-technical teams (sales, marketing, legal, operations) will build their own automation and tools. At Anthropic, a lawyer with no coding experience built self-service tools using Claude Code, reducing marketing review turnaround from 2-3 days to 24 hours

YouTube has overtaken Reddit as the leading source for citations in LLMs.

---

Paid subscribers get the full DoP Substack including: The Knowledge Series for sharpening your tech / AI skills, the AI Prompt and Skills library and DoP Deep dive reports for in-depth analysis to learn lessons from the world’s top tech companies.