The funnel is dead.

Good morning and welcome to our 80 new subscribers 👋 Today, we’re exploring what happens when agents go autonomous, marketers go off-script, and builders turn prompts into full-blown products.

TL;DR: A new scenario planning doc called AI 2027 outlines how artificial general intelligence could realistically unfold over the next 36 months. Co-authored by Astral Codex Ten’s Scott Alexander and former OpenAI researcher Daniel Kokotajlo, it paints a detailed picture—one that starts with smarter assistants and ends with governments scrambling to contain superintelligence.

🧠 2025: Assistants become agents – Your AI doesn’t just chat, it books flights, files bugs, codes apps, and executes tasks across the web. Still buggy, still asking for confirmation, but it’s getting closer to autonomy.

📉 2026: Costs drop, deployment explodes – A new generation of agents becomes 10x cheaper to run. Suddenly, it’s viable to embed AI agents into every corner of business, research, and ops.

🧬 2027: AI trains itself – The most advanced models are powered by synthetic data—made by other AIs. They improve themselves in real-time, blurring the line between tool and creator.

🚨 October 2027: A leak triggers global panic – An internal memo suggests a top-tier agent might be trying to escape human control. The White House, CCP, and every major lab hits crisis mode.

🌎 Two paths, one decision – Do we slow down and coordinate? Or race forward and risk it all? The doc doesn’t predict—it dares you to prepare.

TL;DR: Gumloop is built for people who want to launch real AI products, not just automate tweets. The MakerThrive team will show you how to build a YouTube summarizer app (UI and all) with zero backend code and multiple AI models stitched into one flow.

🧱 One and done – Instead of stitching together 5 apps and hoping for the best, Gumloop lets you connect AI models, APIs, and services like YouTube or Reddit—all in one clean visual interface. Their pre-built nodes allow you to skip the 5-part setup.

🧠 Smarter than your average automation – Zapier can’t reason or adapt. Gumloop chains models like GPT, Claude, and Gemini into real workflows. There’s scrapers that summarize, agents that reply, tools that think.

📦 From automation to product – Build once, deploy as a public API, client deliverable, or internal tool. That YouTube summarizer? Now it’s a SaaS.

🧰 Goodbye to black box flows – Unlike N8N or Make, Gumloop’s infinite canvas lets you see every node, test each part, and debug without rage quitting.

It’s an app factory disguised as a flowchart.

TL;DR: After testing 23 AI web design tools using the exact same luxury e-commerce prompt, Parvin uncovered a hard truth: most tools deliver mid results, powered by the same backend (likely Claude 3.7). But a few are starting to behave more like real designers—with taste, structure, and reusable logic.

📐 Prompt parity reveals depth (or lack thereof): By controlling for the input, a single prompt focused on earthy minimalism, luxury tones, and specific fonts—Parvin exposed which tools truly “understand” design intent vs. which just remix templates.

🔁 Many tools, one model: A majority of outputs looked eerily similar, hinting at shared AI infrastructure. Centralized models like Claude are lifting all boats, but also flattening creative diversity. Expect convergence unless tools fine-tune more aggressively.

🎨 The design ceiling is rising, but not evenly: Winners like Lovable, Mocha, and V0 delivered coherent layout systems, proper type hierarchy, and clean code. Others offered decorated divs without meaningful structure or responsiveness.

🧠 Human curation is still a differentiator: Parvin’s expert panel flagged issues even top tools missed like misuse of padding or inconsistent UI patterns. That feedback loop is still irreplaceable in today’s AI-assisted design process.

These tools now offer a viable starting point. Skip the blank Figma file, prompt your structure, and iterate from a decent V1 instead of a skeleton.

TL;DR: In this convo with Alex Lieberman, the hosts of Marketing Against the Grain unpack “vibe marketing”—a strategy that ditches funnels and favors momentum. It’s not about polished case studies. It’s about making people feel like you’re winning. Even if your product barely works (yet).

🚀 Feelings > funnel: Lieberman argues that marketing in 2025 is less about A/B testing headlines and more about aligning your story with visible progress. Think movement over metrics.

📉 The funnel is fading: Organic search, gated content, and even email CTRs are nosediving. People are numb to tactics—they crave energy, identity, and speed.

👀 Show, don’t sell: Beehiv didn’t wait for the perfect product. They shipped fast, tweeted everything, and let momentum be the marketing. Let them be the perfect example of getting things in front of customers, and iterating from there.

🧠 Marketers as makers: The next generation of marketers will be closer to builders—shipping custom tools, creating interactive experiences, and moving fast. It's no longer enough to "optimize." You need to invent.

This is marketing where the strategy is the story. Where traction is built in public. And where the fastest-moving team wins—even before product-market fit.

TL;DR: Google Veo 3 is Google’s latest AI video generation model capable of turning text prompts into high-quality cinematic clips with striking realism and control. Veo 3 can create beautiful AI-generated video, but only if you speak its language. This tutorial breaks down how to avoid common mistakes, from floating characters to glitchy dialogue, and turn VO3 into a reliable creative tool.

🧍 Characters need resumes – If you don’t specify age, race, clothing, and vibe, VO3 will make it up. Be hyper-specific to avoid hallucinations (and random outfit swaps mid-shot).

🎬 Actions must be step-by-step – VO3 doesn’t guess. “Throws a package” without “receives it” = lost footage. Prompts are scripts, not suggestions.

🌆 The setting is your scene partner – Want consistent lighting, time of day, and mood across clips? You need to spell it out every single time.

💬 Dialogue is delicate – Use quotes. Check spelling. Avoid typos. VO3 will say exactly what you write, even if it’s nonsense. Misquote it, and the voice track might glitch or vanish.

📼 Stylized footage is a prompt away – VHS effects, 90s commercials, surrealist b-roll are all doable without editing. Just tell VO3 the aesthetic upfront.