Build Signals Roundup #2

Hi there,

This past week, I've been using Lovable to build out an idea for a user testing product I've had. There's been such a lot of M&A activity in this space that it's got even harder to just get 5 user tests done! That's the problem I'm planning on solving with this MVP and I hope to share a bit more about it in the coming weeks. If anything, it's proof that the builder economy is there and the gap from idea to launched in the market is slimmer than ever.

It's also been an exciting few days in AI, so much so, it's been hard to keep up!

Again, if you want a deeper analysis and my personal take on how these shifts in tooling, AI systems and product design are reshaping how we work - you should sign-up to my personal newsletter Build Signals.

Here’s what you can expect this week:

📎 Five curated links worth your time

💬 My take on why each signal matters

💡 Practical ideas you can build on right away

I’m building this for people like you. Builders, operators and thinkers who want to create work that matters.

Oh, and don't miss my upcoming event where you can start building your next idea within 60 minutes!

Let’s dive into this week’s edition 👇

1. AI Workflows, Agents, and MCP - What They Are (and When to Use Them)

Everyone’s talking about AI agents, but most people are confusing three very different things:

1. AI workflows
2. AI agents
3. MCP (Model Context Protocol).

This post breaks down what each one actually does, how they relate and when you should be using which.

If you’re experimenting with AI tools, or trying to embed them in product or ops, understanding this is key.

My take:

• AI workflows are the new Zapier. Rule-based, predictable, and super useful, but not intelligent.
• AI agents are where it gets interesting - they reason and act on intent, not just triggers.
• MCP is the enabler. It’s what lets your AI do things, not just chat.
• Most people don’t need to build with MCP - but they should understand its role.
• Knowing which level you’re operating at helps you stop building spaghetti 🍝 and start designing real systems.

AI agents without workflows are chaotic. Workflows without agents are basic. The best teams will learn to combine both and build on MCP when they need scale.

2. Clay Isn’t Just for Lead Lists — It’s a Growth Ops Power Tool

One for the business development folks! Yes, Clay can automate lead sourcing and enrichment. But that’s just scratching the surface.

This post breaks down how smart teams are using Clay for trigger-based prospecting, TAM analysis, event targeting and even pipeline attribution, turning it into a true growth engine.

It’s a blueprint for scaling outbound strategically - way better than just sending out emails and hoping something sticks.

We've never had it this good!

📌 Benchmarks to aim for:

• 5–15% response rates
• 20–30% meeting conversions
• 5–10x ROI on outbound

3. Free Event – UX x AI: What’s Changing and What Still Matters

🗓️ My first event in a long time!

I’m hosting a free, fast-paced session on how AI is reshaping UX and what smart product teams are doing to keep up. This is for designers, PMs and anyone building with AI in the loop.

We’ll cover:

• What “good UX” even means in AI-native products
• Where AI is breaking traditional design patterns
• How to design better prompts, flows and feedback loops
• What still matters when AI tries to do the thinking

Details:

🗓 Wednesday, April 23 at 5:30pm UK Time

📍 Online + Free

💬 Come with questions

4. Llama 4 Launches - Meta’s Open-Source Power Play in Multimodal AI

Meta just dropped its most advanced suite of open-source models yet - Llama 4 Scout, Maverick, and a preview of Behemoth - ushering in what they’re calling “a new era” for the Llama ecosystem.

Here’s the signal:

• Scout is tuned for tasks like reasoning and summarisation, and outperforms Gemini 2 and Google’s Gemma 3
• Maverick is positioned as the top-tier multimodal assistant, beating GPT-4o in Meta’s own benchmarks
• Behemoth (still in dev) will act as a “teacher model” with state-of-the-art benchmarks in math, 'multilinguality' and image processing
• All models use Mixture of Experts (MoE) architecture - smarter routing, less compute, better performance
• Scout + Maverick are openly available now via Meta’s partners and platforms like Hugging Face

My take:

• Meta’s clearly going all-in on open-source AI dominance — and they’re getting serious about multimodal
• MoE is more than just efficient — it’s a sign that we’re building systems that reason modularly
• If Llama continues to outperform, expect real product teams to shift away from closed models
• This is also a design challenge: How do you interface with models that reason across text, images, audio, and video?
• Meta’s long game is clear — build the best base models + get them into the apps people use daily (WhatsApp, Insta, Messenger)

Is this the most serious open-source threat to closed foundation models we’ve seen? 👀

5. Google’s Agentspace + A2A: AI Agents Are About to Get Collaborative

Google recently announced Agentspace, a new platform that lets AI agents securely connect to your most-used enterprise tools.

Think of it as a command center for intelligent workflows. Alongside it, they launched Agent2Agent (A2A), an open protocol designed to let agents from different vendors and frameworks communicate, coordinate and collaborate across enterprise systems.

My take:

• Agents are evolving from individual taskbots to networked collaborators
• A2A could be the API moment for agents
• Google’s playing to its strengths here: infrastructure, scale and enterprise-grade trust
• This opens up real B2B use cases - think agent-to-agent handoffs between departments, tools and even companies
• Big question: Who owns the coordination layer? Whoever wins that builds the new OS for work

This is a massive unlock. Once agents can talk to each other, they negotiate, delegate and orchestrate. That’s new!

If this roundup hit home, you’ll probably enjoy the deeper dives I share in my personal newsletter, also called Build Signals.

Every week, I send one sharp insight, a few curated signals and the strategy behind what matters in product, AI and growth. As well as that, starting to share some of my personal journey too!

👉 Subscribe here to get it straight to your inbox.

Thanks,

Ross Chapman