Microsoft & OpenAI Rethink Their $13B Deal

Plus: Slack’s 2-minute breach response & Google’s push for private AI

Hello!

Innovation gets the spotlight, but real progress is about trade-offs.

Do you automate processes or stick with manual control? Accept “good enough” engineering or push for higher standards? While we’ll try to answer them, there’s some trouble brewing on the horizon.

OpenAI and Microsoft are renegotiating the future of their partnership, and Atlassian is forcing customers into the cloud on its timeline, not theirs

On a more positive note, privacy and security are becoming more important. Slack’s new feature will keep you safer than before, and Google’s VaultGemma model is a step in the direction of private-by-design AI.

And if you’re looking for a palate cleanser, check how people around the world are using ChatGPT and Claude.

Grab your coffee, relax, and enjoy Frictionless!

In the Queue

Reduce Friction

A simple line graph titled “Elliot’s Commit History” with x-axis from 2010 to 2025 and y-axis showing lines of code written. From 2010–2012: Steep rise in commits. 2012–2015: Decline as Elliot became a manager and later director. 2015–2025: Very low activity. Around 2025: Sharp spike upward marked “I started vibing,” representing a surge in commits after adopting vibe coding with AI.

Source: Elliot Graebert

Agentic AI Has Changed My Career

I’m normally sceptical about vibe coding, but it’s still a great tool for non-devs dealing with coding issues. For example, Elliot went from barely coding at all to being a top contributor by pairing “vibe coding” with agentic AI. Adding in agents across tools and workflows helped him hit 10x productivity.

Just Enough Automation

Automation works best when tasks repeat often, effort is low, or the long-term payoff is clear. The best thing you can do is to treat automation like an investment: weigh the effort, costs, and benefits before jumping in.

Solve Customer Problems

"Customers don’t buy products; they hire them to do a job." That’s the secret behind companies like Airbnb, Amazon, or Slack. Instead of chasing personas or quirky campaigns, they focused on becoming indispensable to their customers, and so should you.

The Hidden Cost of “Good Enough”

“Good enough” is how great teams become mediocre. Instead of settling on the new low, push back. Reward high standards, enforce basics, and set the foundation so AI brought into the project won’t multiply the team’s mistakes.

Deepen Your Expertise

A horizontal timeline with four colored boxes marking Atlassian’s Data Center phase-out: 2025 (blue): No new Data Center app submissions (Dec 16, 2025). 2026 (green): End of license sales to new customers (Mar 30, 2026). 2028 (orange): End of license sales to existing customers (Mar 30, 2028). 2029 (purple): End of life of Data Center products (Mar 28, 2029).

Source: Atlassian

Atlassian's Move to Cloud-Only Means Customers Face Integration Issues and More

By 2029, Atlassian will close off its datacenter products, pushing everyone to the cloud. They promise savings and shiny AI features, but once add-ons are factored in, many enterprises will pay ~28% more. Atlassian might say its pro-customer values influenced this decision, but users don’t seem happy with it.

Helly Hansen Migrated to Vercel and Drove 80% Black Friday Growth

Helly Hansen ditched Adobe Commerce PWA for Next.js and Vercel, and the payoff was huge: faster pages, stable Black Fridays, and 80% YoY growth with zero downtime. The switch fixed their site’s performance and gave them more room to experiment.

Best CMS in 2025: Sanity vs WordPress

WordPress is still the fastest way to spin up a site, thanks to plugins and themes. But Sanity offers structured data, multi-channel publishing, and performance at scale, so which one to pick? If you want a quick setup, go with WordPress. If you’re building for growth, Sanity is hard to beat.

Building Slack’s Anomaly Event Response

Cybercrime is on the rise, and Slack decided to step up to the challenge. Their new Anomaly Event Response system shuts down risky sessions the moment it detects suspicious behavior. Breaches that once took days to contain can now be stopped in minutes.

AI Corner

A grid of logos (GitHub, Google, Dropbox, Microsoft, Monzo, Atlassian, Adyen, Booking.com, Grammarly) with bullet points under each showing how they measure AI’s impact. Metrics include things like PR throughput, cycle time, AI time savings, developer satisfaction, acceptance rates, and AI spend.

Source: Pragmatic Engineer

How Tech Companies Measure the Impact of AI on Software Development

If you think GitHub, Google, or eBay evaluate AI impact by counting likes, think again. Dropbox, for instance, measures PRs from AI users or the AI spent per dev. Here’s how 18 top tech companies in the world measure the impact AI has on their business. 

OpenAI and Microsoft Sign Preliminary Deal to Revise Partnership Terms

Just last week, we were wondering if the partnership between OpenAI and Microsoft would last. Turns out, both companies signed a deal to revise it, as OpenAI shifts to a public benefit corporation. They want more compute and new partners; Microsoft wants to protect its $13B stake. What’s next? 

VaultGemma by Google: The World's Most Capable Differentially Private LLM

Privacy doesn’t have to come at the cost of utility. Google’s new VaultGemma is the largest open model trained to keep your data safe. Its performance is comparable with models from five years ago, but VaulGemma proves that private-by-design AI is possible.

Anthropic: Writing Effective Tools for Agents — With Agents

Agents only work as well as the tools you give them. Anthropic shares how they prototype, evaluate, and even use Claude to optimize tool performance. Their advice is to keep tools focused, efficient, and intuitive for agents and humans alike.

Just Cool

Four scatter plots showing how different job groups use Claude, with bubble size representing usage volume: Computer & Mathematical: No clear correlation; wide spread of usage. Educational Instruction & Library: Mostly clustered at low usage. Arts, Design, Entertainment, Sports, Media: Slight upward trend, some higher adoption. Office & Administrative Support: Stronger correlation; higher adoption linked to higher AI Usage Index. Bubbles are shaded by adoption tier (minimal to leading).

Source: Anthropic

How People Actually Use ChatGPT Vs Claude - And What the Differences Tell Us

OpenAI and Anthropic published their research on how people use their models, and their roles couldn’t be more different. ChatGPT is busy ghostwriting emails and answering random life questions, while Claude is off cranking through code on autopilot.

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.