AI Slows Dev Productivity by 19%

Kimi K2 DeepSeek Moment, AI Affects Speed & Kiro IDE Debut

Happy Wednesday!

AI was supposed to save us time, but a new study shows developers are working 19% longer with AI tools. Maybe they just haven’t tried Kiro, AWS’s new IDE perfect for those tired of vibe coding.

Meanwhile, Moonshot’s Kimi K2, backed by Alibaba, has entered the AI race, reportedly outperforming Claude and GPT‑4.1 in coding tasks. Next.js 15.4 is out too, with a promising preview of what version 16 might bring.

I also rewatched Malte Ubl’s keynote from Vercel Ship last week. It inspired me to describe Vercel’s three-phase approach to AI agent development, which you can read here.

Now, grab your coffee, relax, and enjoy this week’s issue of Frictionless.

In the Queue

Reduce Friction

Illustration of a cityscape labeled “Startups” under a rain cloud marked “Cloud” with logos of Meta, Google, and Microsoft. The cloud rains down “Money,” while “Team” floats upward like sunshine from the startups to the cloud. Visual metaphor for how big tech funds startups but also absorbs their best talent—“Hire And License Out” model.

Source: kwokchain

The HALO Effect

A new model is taking shape in Silicon Valley: hire great people, build great models, and then license them out.  The “Hire and License Out” model is already used by companies like Inflection, Character AI, and Windsurf. Let's look at what makes it so compelling.

Speed Without Quality Is Tomorrow’s Crisis Arriving Faster

We’ve all felt the pressure of deadlines, but at Epic Test Quest, skipping testing cost €100,000 in revenue. The lesson here is simple: speed might be important, but without quality, it becomes a liability.

Four-panel comic where a developer warns their code is messy. The reviewer reacts humorously: describing it as a house built with a hatchet, a salad recipe written by a lawyer using Excel, and finally a chaotic IKEA argument transcript that compiles. Ends with the developer agreeing to read a style guide.

Source: xkcd.com

Checklist for a Perfect React Code Review 

Code reviews are the backbone of any dev team. Their real value is in better communication, shared context, and dev-to-dev mentorship. If you want to get 100% out of your reviews, this checklist is a good place to start.

How to Talk About Your Competition

Having competitors means you're in a healthy market. However, standing out takes more than just a better product. You also need to know how to talk about the competition without sounding insecure or petty.

Deepen Your Expertise

Technical diagram labeled “API Gateway High Level Design.” It shows gateway clients routing requests through an API gateway, which applies configuration logic and filters—both pre- and post-processing—before sending them to backend internal services. Boxes represent components like route config, handlers, predicates, filters, and services, laid out in a left-to-right flow.

Source: ByteByteGo

How Tinder’s API Gateway Handles A Billion Swipes Per Day

Tinder’s API gateway handles traffic from 190+ countries, routes calls across 500+ microservices, and fends off bots at scale, all with low response times. Their secret? A custom gateway built in-house to scale fast and stay secure.

Figma's $300,000 Daily AWS Bill Highlights Cloud Dependency Risks

By spending 12% of its total revenue on AWS, Figma became fully dependent on a single provider. If Amazon changes its terms or pulls support, there’s no easy way out. Even top-tier companies can get trapped by scale, which makes now a good time to rethink your infra strategy.

Next.js 15.4

The new update to Next.js focuses mostly on stability and Turbopack. What takes the spotlight, though, is the preview of the upcoming Next.js 16 features. There isn’t a concrete date yet, but we might expect a release this summer.

To Be a Better Programmer, Write Little Proofs in Your Head

Sketching “little proofs” in your head, preconditions, invariants, and all, can make your code more correct on the first try. It’s a simple habit that can make a big difference for your team.

AI Corner

Bar chart showing average time forecasts versus actual implementation times for developers using AI tools versus those who don’t. AI-assisted developers predicted shorter completion times (0.76x), but in reality took longer than expected (1.34x). The chart highlights discrepancies between AI expectations and observed outcomes with confidence intervals.

Source: Joel Becker, Nate Rush, Elizabeth Barnes, David Rein

Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity

A new study shows something unexpected: experienced open-source devs using AI tools were 19% slower. One theory points to the steep learning curve of AI tools as the reason; another suggests that they spent more time cleaning up the messy code AI produced, but what is the truth?

Context Engineering Guide

Prompt engineering isn’t enough anymore. If you want to stay up to date with LLMs this guide explains what context engineering is with practical examples of prompt design, memory handling, and building effective agent workflows.

Bar chart dashboards comparing the performance of different AI models across five benchmarks: Humanity’s Last Exam (HLE), xbench-DeepSearch, FRAMES, SEAL-0, and SimpleQA. Kimi-Researcher consistently outperforms models like OpenAI, Gemini, DeepSeek, Claude, and Grok across all tasks. Metrics are shown in percentage scores or success rates, with Kimi bars highlighted in blue for visual contrast.

Source: Moonshot Tech Blog

Alibaba-Backed Moonshot Releases New Kimi AI Model That Beats ChatGPT, Claude In Coding — And It Costs Less

Another DeepSeek moment? Moonshot AI, backed by Alibaba, released Kimi K2, an open-source model trained on 15.5 trillion tokens. It reportedly rivals Claude and GPT-4.1 in coding tasks, at a fraction of the cost.

AWS Previews Kiro IDE for Developers Who Are Over Vibe Coding

After the recent Cursor controversy, AWS unveiled Kiro, its new AI-powered IDE. Unlike some of its competitors, Kiro doesn’t rely on vague prompts. It starts with structured specs and guides you through requirements, design, and deployment.

Just Cool

Timeline visual (not directly viewable here but inferred from the filename and typical content) illustrating the evolution of React through significant codebase changes. Likely uses branching arrows, version numbers, and small code snippets to show how React’s architecture and features have shifted over time.

Source: Playful Programming

The History of React Through Code

Has React changed a lot since 2011? Take a look at how React evolved through its code and judge for yourself!

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.