Why 95% AI Projects Fail & Next.js 15.5

Sam Altman on AI Bubble & Forgotten Factor in Estimation

Hello!

There’s no better teacher than failure.

With that in mind, what can we learn from 95% of generative AI pilots failing? That AI itself won’t deliver value. Duolingo CEO found out the hard way after users protested his AI memo a few months back, and now he’s back with an explanation

Maybe the backlash doesn’t come from the right people? While the consumers weren’t enthused about the GPT-5 release, companies like Vercel and Cursor are thrilled. Does this mean a shift towards enterprise in AI?

This week, we’re also looking into the reality of AI-assisted productivity, lessons learned from Netflix’s discovery algorithm, and why communication, not code, is still the biggest factor in hitting delivery dates.

Grab your coffee and enjoy this issue of Frictionless.

In the Queue

Reduce Friction

A cartoon of two cavemen pulling a cart full of heavy rocks with square wheels. Another caveman offers them a round wheel, but they reject it, saying “No thanks! We are too busy.” The cartoon satirizes how people often reject productivity improvements because they’re focused on immediate work.

Source: S.S Smith

The Hidden Cost of Slow Feedback Loops

Every 10-minute delay in verifying code adds up until you’re effectively short one engineer. If your team’s shipping speed is slowing, check the feedback loop first.

The Most Underestimated Factor in Estimation

After 200+ software builds, Pawel Brodzinski says the biggest factor in delivery speed isn’t dev speed, but communication. Fast code won’t save your release if misalignment and rework keep getting in the way.

How a Next.js Migration Grew Revenue 24% for an e-Learning Platform

Learn Squared was stuck with a slow Drupal site that frustrated users. After migrating to Next.js, their load times improved, content management became simpler, and users happier, increasing revenue by 24%.

How to Sync Your Budget with a Strategic Plan

Strategic plans talk vision, budgets talk numbers, and they rarely talk to each other. If you're planning for next year, ground both in stakeholder priorities to turn strategy into action.

Deepen Your Expertise

Two heatmaps side by side. The left shows a sparse user-item rating matrix with users (Alice, Bob, Charlie, David, Eve) rating movies in different genres, with a color scale from 2.0 to 5.0. The right shows a dense user-feature latent factor matrix, visualizing compressed preferences in red-blue shading. It illustrates how Netflix turns sparse ratings into latent features to make recommendations.

Source: Beyond IT

Inside Netflix’s $1 Billion Algorithm

Discover the matrix math behind Netflix’s recommendation engine, how real-time learning and A/B testing help with product discovery

5 Best Headless CMS Platforms in 2025

Picking the best headless CMS can feel like a maze, so we’ve prepared a review of 5 best CMS solutions in 2025. Find out where Sanity, Strapi, Storyblok, Prismic, and Contentful shine, and when you’re better off with another option.

Next.js 15.5

We might be waiting a bit longer for Next.js 16, but 15.5 is already paving the way with smarter TypeScript, stable middleware, and production turbopack builds. Best start prepping now before the deprecations hit.

Server and Client Component Composition in Practice

Working with RSCs in Next.js? Learn how to properly compose client and server components to avoid unnecessary re-renders, and reduce bundle size, to keep your app performant.

AI Corner

A bar chart titled “The steep drop from pilots to production for task-specific GenAI tools reveals the GenAI divide.” General-purpose LLMs: 80% investigated, 50% piloted, 40% successfully implemented. Task-specific GenAI: 60% investigated, 20% piloted, 5% successfully implemented. It shows a steep decline in adoption as projects move from pilot to production, especially for task-specific tools.

Source: MIT

MIT Report: 95% of Generative AI Pilots at Companies Are Failing

Despite all the hype, only 5% of generative AI pilots are delivering real business value. The rest are stuck because of vague goals and over-investment in the wrong areas.

Sam Altman Says ‘Yes,’ AI Is in a Bubble

Even OpenAI’s CEO is cautioning that today’s AI boom echoes the dot-com bubble. Who will be left standing once it bursts - and who will disappear with it?

The CEO of Duolingo Wants to Have a Conversation About AI

What happens when the world’s most beloved language app says it's going all-in on AI? The CEO is back in the spotlight, trying to prove there's still room for humans at Duolingo.

The Reality of AI-Assisted Software Engineering Productivity

Addy Osmani digs into what the data says about AI coding tools in 2025. While most devs see modest 20–30% gains on greenfield work, debugging and reviewing “almost right” code slows the teams down.

GPT-5’s Rollout Fell Flat for Consumers, but the AI Model Is Gaining Where It Matters Most

Regular users might be unhappy with GPT-5, but that’s not who it was built for. Vercel, JetBrains, and Cursor are already making it their default for coding and automation, drawn by better problem-solving and lower costs than Claude. Looks like enterprise, not consumers, is where AI will focus.

Just Cool

An underwater photograph of a diver swimming among a group of sperm whales. The whales are massive, dark, and surrounding the diver in a calm, blue ocean. The diver is dwarfed in scale, creating a striking image of harmony between humans and marine life.

Source: The Atlantic

Ocean Photographer of the Year 2025 Finalists

Take a breather and dive into the Ocean Photographer of the Year 2025 gallery. From breaching orcas in Puget Sound to a sneezing marine iguana in the Galápagos, see the marine wildlife and its most surprising.

Let’s Stay in Touch! 📨 

Do you have any comments about this newsletter issue or questions you want to ask? Drop me a message or book a meeting.

What do you think of today's email?

Your feedback helps me improve Frictionless.

Login or Subscribe to participate in polls.