Spotify's co-CEO revealed this week that the company's top engineers haven't manually written a line of code since December. Instead, they're using an internal AI system to direct a model to build features and fix bugs, often right from their phones.
This signals a massive shift in the developer role, from writing code line by line to orchestrating an AI to do the building. Companies that successfully make the jump from doer to director will ship products and run experiments much faster than those still doing it the old way.
Topics of the day:
Spotify says its best devs haven't coded since December
OpenAI drops an ultra-fast coding model with 1000+ tokens per second
China's Zhipu AI releases a 744B open-source model rivaling Western leaders
A new HBR study finds AI makes workers busier, not freer
AI-supported screening catches 20% more breast cancers
The Shortlist: OpenClaw creator joins OpenAI, Anthropic's free tier upgrade, Meta's feed tuning, ElevenLabs' emotional voices and Cloudflare's agent-friendly pages
Spotify says top devs haven't coded since December
What's happening: Spotify's co-CEO revealed this week that its top engineers haven't manually written code since December. Instead, they're using an internal AI system to direct Claude Code to build features and fix bugs, often right from their phones.
In practice:
This signals a big shift in the developer role from writing code line by line to orchestrating and directing an AI to do the building.
Automation is reaching a new level of efficiency, with engineers able to fix bugs and deploy new features via Slack during their commute.
This AI-first workflow is a massive growth lever, enabling the accelerated development that helped Spotify ship over 50 new features last year.
Bottom line: The definition of a "technical" employee is evolving from doer to director.
OpenAI drops ultra-fast coding model
What's happening: OpenAI released GPT-5.3-Codex-Spark, an ultra-fast coding model designed for real-time AI collaboration. Running on Cerebras hardware, the new model delivers over 1,000 tokens per second, making AI-assisted programming feel nearly instantaneous.
In practice:
This accelerates the shift from manually writing code to directing AI agents, a trend highlighted by Spotify's claim that its top developers haven't coded since December.
Near-instant feedback loops allow product managers and founders to prototype ideas faster, making small, targeted edits to logic or interfaces on the fly without deep coding knowledge.
Faster iteration means you can run more experiments, whether it's building landing page variations, creating custom analytics scripts, or automating internal workflows for your team.
Bottom line: Model speed has been a major bottleneck for AI coding assistants.
DIY, AI family assistant in WhatsApp
Keeping an overview of how the weekly family show, including pickups, grocery shopping, afterschool activities, birthdays and so on can be a pain to be on top of.
Here's how to turn your family WhatsApp group into something actually useful by adding an AI assistant that manages your shared calendar, translates school emails, and sends daily reminders.
What you're building
A WhatsApp bot that lives in your family group chat. Mention it by name and it responds. It can read and write to a shared Google Calendar, translate messages (handy if your kids' school communicates in a language you're still learning), manage a grocery list, and send proactive reminders like morning briefings and pickup alerts.
What you need
An always-on server (a small VPS works, or a Mac Mini at home)
OpenClaw - the open-source AI agent framework
A WhatsApp number for the bot (could be your own or a cheap prepaid SIM)
A Google Calendar you want to share with the family
An Anthropic API key (for Claude as the AI brain)
How to host it
You need a machine that stays on 24/7. Two good options:
Cloud VPS (like Hetzner) - Cheapest route. A 4 EUR/month server handles this easily. You get reliability and remote access, but no access to native apps.
Mac Mini at home - More expensive upfront, but it can run macOS apps. This means your agent could potentially access iMessage, Apple Reminders, or other native apps alongside WhatsApp. If your family is in the Apple ecosystem, this is the more powerful option.
How to set it up
Spin up your server and install OpenClaw following their quickstart guide.
Connect WhatsApp using the Baileys library (open-source, no official API needed). OpenClaw has a WhatsApp integration ready to go.
Hook up Google Calendar API so the bot can read and write events to your shared family calendar.
Configure the agent's personality and rules - tell it which calendar to use, what language to expect from school, when to send morning briefings.
Add the bot's WhatsApp number to your family group. Done.
What it looks like in practice
7am: The bot posts "Good morning! Today: dentist at 10, school pickup at 3:30, football practice at 5."
You forward the school calendar "Please add all holidays to the family calendar and invite my wife". In my case the calendar is Catalan, the bot translates it and adds any dates to the calendar.
Your partner texts "@fambutler add milk and eggs to the list." Done.
3pm: "Reminder: school pickup in 30 minutes."
The whole setup takes an afternoon. The hardest part is the Google Calendar API credentials, and there are step-by-step guides for that everywhere. Once it's running, you'll wonder how you managed without it.
I plan on adding our online grocery shopping site, so it can analyze what we buy and when, then fill up the basket for us.
China's Zhipu AI challenges Western models
What's happening: Chinese startup Zhipu AI released GLM-5, a 744B-parameter open-source model trained entirely on Huawei chips. It's competing with proprietary models from Google and OpenAI on key benchmarks.
In practice:
Its native Agent Mode can turn a simple prompt into a ready-to-use business proposal or financial report in Word, PDF, or Excel.
You can build custom tools on it for a fraction of the cost of its rivals, creating a low-cost alternative for internal automations.
The model is free to try now, giving you a sandbox to test complex document creation workflows before committing.
Bottom line: The gap between open-source and closed models is closing faster than ever. This gives you more leverage and optionality when choosing a foundational model to build on.
Study: AI makes you busier, not lighter
What's happening: A new Harvard Business Review study found that instead of giving us more free time, AI tools often make work more intense. Professionals are expanding their job scope and blurring work-life boundaries because AI makes it feel easier to just "do more."
In practice:
Watch for scope creep, where AI encourages you to take on tasks outside your expertise. Product managers start writing code. Researchers take on engineering work. Roles blur quietly.
Set clear team norms around AI usage to protect downtime, like "no prompting after 6 PM," which prevents work from seeping into every spare minute.
Use AI for batch processing tasks like summarizing meeting notes overnight, rather than creating more multitasking during the day that drains your focus.
Bottom line: The initial productivity rush from AI can mask a hidden cost in burnout and cognitive load. The key is building an intentional "AI practice" with clear rules for how and when you use these tools.
AI boosts cancer detection by 20%
What's happening: A landmark trial published in The Lancet found that using AI as a "second reader" for radiologists increased breast cancer detection rates by 20%. The system flags areas on mammograms that human experts might miss, acting as a highly effective safety net across 579,000+ women.
In practice:
This is a real-world example of AI working as a "second set of eyes" to augment, not automate, an expert's judgment.
You can apply this model to any quality control process, from reviewing legal contracts for missed clauses to checking software code for subtle bugs.
Instead of threatening jobs, this approach enhances human workers. In fact, demand for radiologists is actually growing.
Bottom line: The true power of AI today lies in making skilled professionals better at their jobs. This "expert assist" model is a practical playbook for driving efficiency in almost any field.
What I read/use this week
Tools, articles, and people worth your attention.
RentAHuman.ai - A marketplace where you can hire humans specifically to work alongside AI agents. Your agent needs a phone call made or a physical task done? Rent a human. It's the most literal answer to "what jobs will AI create."
Conductor - A Mac app that lets you run a team of coding agents in parallel. Think of it as a project manager for AI developers, each working on a different part of your codebase simultaneously.
OpenAI Codex - OpenAI's dedicated coding partner that runs in a sandboxed cloud environment. It can handle multi-file edits, write tests, and submit pull requests, all asynchronously while you work on something else.
Context is King by Evan Armstrong - A sharp framework for understanding where AI value actually accretes. Armstrong argues SaaS is splitting into three layers: the model layer, the context layer (where your data and workflows live), and the interface layer.
Cerebras - The hardware company behind Codex Spark's 1000+ tokens per second. Their Wafer Scale Engine 3 chip is purpose-built for inference speed, and the new multi-year deal with OpenAI (reportedly worth up to $10B) signals where the industry is heading.
The Shortlist
Anthropic unlocked file creation, Connectors, and Skills for all free Claude users, giving everyone access to powerful features previously reserved for paid subscribers.
Meta launched a "Dear Algo" feature on Threads that lets you temporarily tune the feed just by typing what you want to see more or less of.
ElevenLabs shipped Expressive Mode for its voice agents, adding real-time emotional range and tonal shifts to make conversational AI feel more human.
Cloudflare introduced Markdown for Agents, a feature that automatically converts web pages into a token-efficient format for AI crawlers, cutting data usage by over 80%.
This newsletter is where I (Kwadwo) share products, articles, and links that I find useful and interesting, mostly around AI. I focus on tools and solutions that bring real value to people in everyday jobs, not just tech insiders.
