Let’s get to today’s topics:

1. New trend: token spend breaks budgets – what next? In the past 2-3 months, spending on AI agents has exploded at many tech companies, and the ramifications of this are starting to dawn on engineering leaders. We’ve sourced details from 15 companies, including the different ways they are coping with this realization.

2. New trend: more AI vendors can’t keep up with demand. Related to massively increased spending, GitHub Copilot and Anthropic are starting to limit less-profitable individual users, so they can serve business users whose spend has easily 10x’d in the last few months. The exception is OpenAI and Codex.

3. Morale at Meta hits all-time low? Business is booming but devs at Meta are furious and worried due to looming layoffs, and an invasive tracking program rolled out to all US employees.

1. New trend: token spend breaks budgets – what next?

Read more

Today, we cover:

1. Tokenmaxxing: weird new trend. At Meta, Microsoft, Salesforce and other large companies, devs are purposefully burning tokens (and money!) to inflate their AI usage and hit AI usage metrics which they treat as targets.

2. Are coding AI-agent subsidies doomed? At the same time as Anthropic stopped subsidizing enterprise plans, Uber managed to burn through its entire 2026 AI token budget in just 3 months. I expect per-engineer AI budgets to be rolled out across more companies soon.

3. Industry Pulse. The myth of Claude Mythos, Claude’s degradation, Cal.com going open source due to AI threat, Vercel open sources its “agent factories” tool, sensible AI usage guidelines in the Linux kernel, and more.

4. Cal.com goes closed source – but is it really because of AI? The open source Calendly alternative moved a good part of its code to a closed repo, citing AI and security concerns. But perhaps this was just a business model change that would have happened, AI or not.

1. Tokenmaxxing: weird new trend

Read more

Today, we cover:

1. Founders back coding with AI: Mark Zuckerberg & Garry Tan. The Meta chief is shipping diffs after 20 years, while Garry Tan at Y Combinator is knee-deep in coding, 15 years later. Founders with technical backgrounds being hands-on with AI agents could be a good thing – especially when the “honeymoon” period ends.

2. A bad week for Claude Code and GitHub. Claude Code’s source code was leaked when a sourcemap file was accidentally uploaded, and revealed that the tool uses anti-distillation to deal with competitors, and also some potential future features such as an always-on background agent. Also: DMCA copyright strikes from Anthropic raise a big question: can a codebase that is fully AI-generated be covered by copyright?

3. Industry pulse. Meta sets targets for AI-generated code, GitHub’s 6 years of reliability issues, massive job losses at Oracle, GitHub Copilot rollouts and then rolls back ads, RAM prices fall (for now), and more.

1. Founders back coding with AI: Mark Zuckerberg & Garry Tan

Two interesting stories of AI tools encouraging busy founders to start writing code again with AI agents.

Mark Zuckerberg back to landing diffs, 20 years later

Read more

Today, we cover:

1. Does GitHub still merit “top git platform for AI-native development” status? Availability has dropped to one nine (~90% – !!), partly due to not being able to handle increased traffic from AI coding agents. There’s also no CEO and an apparent lack of direction.

2. Should a tool auto-add itself as a contributor to PRs? Claude Code and GitHub Copilot auto-add themselves to commits, which is effectively free advertising. Codex and OpenCode purposely do not.

3. Microsoft promises Windows will not be “Microslop.” After years of forced Copilot integrations, Start menu ads, and mandatory Microsoft accounts, the Windows team is promising to undo the self-inflicted damage done to the OS. It’s better late than never, but why did Microsoft allow the “Microslop” perception to stick around so long?

4. Industry pulse. Massive LLM supply chain attack via LiteLLM, backlash after Cursor forgets to mention that Composer 2 is based on an open source model, what happens when you stop reviewing AI code, OpenAI kills Sora, and more.

1. Does GitHub still merit “top git platform for AI-native development” status?

We’re used to highly reliable systems which target four-nines of availability (99.99%, meaning about 52 minutes of downtime per year), and for it to be embarrassing to barely hit three nines (around 9 hours of downtime per year.) And yet, in the past month, GitHub’s reliability is down to one nine!

Here’s data from the third-party, “missing GitHub status page”, which was built after GitHub stopped updating its own status page due to terrible availability. Recently, things have looked poor:

Read more

Today, we cover:

1. Is the FDE role becoming less desirable? Job postings for Forward Deployed Engineers (FDEs) have surged, but many professionals don’t want the role because it’s more like solutions…

Read more

Before we start, I’d like to share updates about data in the two most recent articles:

Uber’s AI adoption numbers. The Dev Platform folks at Uber have been kind enough to share the latest up-to-date numbers on AI adoption, following Tuesday’s article, which reported that 31% of all code is AI-authored. It turns out this was incorrect, due to a bug with one of the tools. Here’s how things look there:

• 84% of devs at Uber are agentic coding users (either CLI-based agents, or making more agentic requests than tab-completion in the IDE).

• 65-72% of code is AI-generated inside IDE-based tools. For AI command line tools like Claude Code the figure is, naturally, 100%.

• Claude Code usage almost doubled in 3 months; from 32% last December, to 63% by February. Meanwhile, IDE-based tool usage (Cursor, IntelliJ) have plateaued.

Separately, last week’s edition of The Pulse reported that Block did not make job cuts between 2022 and 2025, which was incorrect. Layoffs happened in Jan 2024 and March 2025. I have updated my analysis with these details; apologies for the error.

Today, we cover:

1. Staff+ engineers in 2027 and beyond. What happens to the Staff engineer role when agents write more code? Actually, they could be more in demand than ever!

2. New trend? AI token costs are a rising concern for CTOs. Accounts from two engineering leaders who are raising the alarm about steeply climbing AI costs and the need to slow down spending.

3. 10% layoffs at Atlassian: is it AI’s fault? Atlassian says it wants to invest savings in AI, but is there more to it?

4. Industry Pulse. An AI-powered library reimplementation sparks copyleft licensing debate, Anthropic launches $15–25 per-review code reviews, Microsoft ships a Claude-powered, Copilot Cowork clone, and Apple is the lone Big Tech not ramping up AI infrastructure spending.

1. Staff+ engineers in 2027 and beyond

At a recent two-day workshop in Utah, US, named ‘The Future of Software Development’, and organized by Martin Fowler, I was among 50 attendees. We self-organized our own sessions, where everyone could suggest a topic close to their heart, which all went on an agenda:

Read more

Today, we cover:

1. Military strikes take down cloud services for the first time. Drone attacks in Bahrain and the UAE took AWS datacenters partially or fully offline during the early days of the Iran…

Read more

There are several layers to dig into h…

Read more

Read more

Today, we cover:

1. New trend: open source projects overwhelmed by AI-generated security reports. Projects like Node.js, Django, and Fastify are restricting or dropping vulnerability reporting platforms like HackerOne.

2. AI agents keep changing engineers’ working styles. Four cases of tech professionals changing how they work thanks to AI agents – including Uncle Bob Martin (author of “Clean Code”) who might entertain the possibility that code readability matters less with AI.

3. Layoffs at big, profitable companies. Amazon cuts 16,000 corporate jobs and Pinterest lays off 15% of staff, despite the businesses being profitable. Both sets of job losses seem “quarterly-driven” and timed before earnings announcements.

4. “AI-native” techies increasingly in-demand: two stories. A product manager who spent 2,500 hours building with AI agents finally landed his dream engineering job specifically because he’s “AI-native.” Meanwhile, a startup failed to hire a junior AI engineer in two months of searching.

5. Industry Pulse. Claude Code installs exploded in January, OpenAI acqui-hired most of the Cline team, Anthropic forces Clawdbot rebrand, GitHub is finally fixing its slow UI, China’s Kimi K2.5 matches Opus 4.5 for a fraction of the cost, and one AI code review vendor questions whether the space is in a bubble.

Programming note: for the next two weeks, I’ll be travelling in the US, so there will be articles on Tuesdays and new podcast episodes on Wednesdays during that fortnight. The upcoming podcast episode is with the legendary Grady Booch, who offers some much-needed perspective on how the rapid change being wrought by AI agents has actually occurred before, and many software engineers back then were similarly worried that the technological leap would kill the tech industry. Of course, the opposite happened.

1. New trend: open source projects overwhelmed by AI-generated security reports

Read more

Today, we cover:

1. Why the claim “writing code by hand is almost dead” feels so dramatic. It used to take years for major technological shifts in computing, like going from Assembly to high-level languages, to play out. In stark contrast, LLM tooling has gone from “meh” to “good-enough-to-write-most-code” in mere months.

2. New trend: staff+ engineers and VPs making heavy usage of AI tools at tech giant. Microsoft’s internal dashboard for AI usage reveals interesting details, including that senior folks who barely used to code are now the biggest users of AI agents – some of it for coding use cases, others for planning and exploring.

3. Replacing a $120/year micro-SaaS in 20 minutes with an LLM. I used to pay $120/year for a SaaS that hasn’t added new features in four years, and didn’t fix its broken billing system for three years. Using an LLM, I managed to rewrite all the functionality I used to pay for in 20 minutes. Is this bad news for “write once, don’t update later” SaaS?

4. Industry Pulse. Claude Code looks more like an IDE with its new diff view, ChatGPT to host ads while Gemini stays ad-free, how Anthropic does takehome exercises during interviews, Cloudflare acquires Astro, and more.

Before we start: last call for The Pragmatic Summit. Ninety percent of spaces are now filled. If you will be in San Francisco, US, on 11 February, you can apply here. We aim to prioritize paying subscribers like yourself, and recorded sessions will be shared with paid readers first as a thank you for your ongoing support!

1. Why the claim “writing code by hand is almost dead” feels so dramatic

Read more

Today, we cover:

1. Long-running agents, and orchestrating lots of them. Expect projects and experiments which push agents to go further this year, with early examples from Cursor, and the author of “…

Read more

Today, we cover:

1. Feedback on Tuesday’s deepdive. A lot is unknown with more capable AI agents, and after using these better AI agents to get legal advice, I’m also more bullish on software engineer…

Read more

• Check out our new MCP report! Elin and I have done something experimental: create a more detailed report than even our MCP deepdive was. If you’d like to get deeper into the MCP ecosystem, check out the report. Feel free to give feedback, as we’d like to do more of these if there’s enough interest to justify the intensive research involved. This is the most detailed report on this topic that we know about.

• The Pragmatic Summit speaker list is nearly finalized. Speakers confirmed in the past few weeks include some names you may know as guests on the Pragmatic Engineer podcast: Chip Huyen, Martin Fowler, Nicole Forsgren, and Kent Beck. I’ll share the detailed agenda in early January. The event takes place on 11 February, in San Francisco. Apply here to secure a spot.

The Pulse is a series covering events, insights, and trends within Big Tech and startups.

Today, we cover:

1. Internal dev tooling at Meta & “trajectories.” An overview of three different internal AI-coding tools which devs at Meta use. Also, the company has started sharing the prompts which devs make when generating code – and it’s pretty controversial.

2. GitHub upsets devs by charging for self-hosted CI/CD. GitHub Actions is notoriously slow and unreliable. But instead of improving the service, GitHub sought to charge by the minute for using third-party CI/CD solutions. That didn’t go down well with devs.

3. Industry Pulse. Warsaw on track to become the “tech capital of the EU,” hiring juniors could actually be “profitable” thanks to AI tools, GitHub is working on stacked diffs, non-devs use LLMs for more coding tasks, Cursor migrates CMS to Markdown, OpenAI gets rid of 6-month vesting cliff, and more.

4. “Apple Tax” alive in Japan. A US court rejected Apple’s appeal, and the iPhone maker can no longer ban alternative in-app payments or impose junk fees on third-party payments… in the US. Meanwhile in Japan, Apple is pulling the same trick that got it in trouble in the US.

1. Internal dev tooling at Meta & “trajectories”

An interesting detail about Meta that I learnt from chatting with two current software engineers there: last week, the social media giant rolled out a feature allowing devs to see the AI prompt history on diffs (internal lingo for pull requests) in review. And this new feature launch is causing some controversy in the workplace!

Read more

Today, we cover:

1. Cloudflare’s latest outage proves dangers of global configuration changes (again) Deja vu: a large Cloudflare outage caused by an instantly rolled-out global config change – two we…

Read more

Today, we cover:

1. Could a 5-day RTO be around the corner in Big Tech? From next February, workers at Instagram must be in the office, five days a week. This makes Meta the second tech giant after Amazon to mandate a 5-day RTO. Will more big companies do the same?

2. Are senior engineers better with AI than juniors are? Data from more than 40,000 users of Cursor suggests that experienced devs get better results from AI agents. It suggests the tool amplifies coding skills – not replace them.

3. Industry Pulse. OpenAI dumps Mixpanel after security breach, Stripe buys usage metering startup, OpenAI’s “Code Red,” a high-risk security issue at a $3B legal startup, and more.

4. Anthropic and Claude Code on a roll. Claude Code just surpassed $1B in annual run rate revenue after 6 months, Anthropic buys JavaScript runtime Bun, and might plan to go public as soon as next year.

1. Could a 5-day RTO be around the corner for Big Tech?

Read more

11 February 2026, in San Francisco – I hope to see many of you there! You can also apply here directly.

Read more

Today, we cover:

1. Microsoft too early to agentic OS – like with Windows phones? The tech giant wants Windows to be an “agentic OS”, but developers hate the idea and could move to MacOS or Linux. Is Microsoft in the process of repeating its Windows Mobile 6 error?

2. Industry pulse. Agents becoming the default way to write code, more AI-coding startups adopt metered pricing, Cursor the highest-valued dev tools company, 1-day onboarding at AI-native companies, and more.

3. Inside Cursor’s unique engineering culture. A completely different way of hiring, rapid shipping as the norm, many former founders at the company, and more.

Before we start: as a paid newsletter subscriber, you can apply a few days early to The Pragmatic Summit!

I shared a teaser on Tuesday: and I’m now very excited to share the first major in-person The Pragmatic Engineer event. This will be a gathering of 400 top engineers and leaders in San Francisco on 11 February 2026 for a day of practical insights, high-signal sessions, and peer exchange focused on building great products and scaling world-class engineering teams.

Read more

Today, we cover:

1. Cursor and GitHub double down on agents. Each company is focusing on agents: Cursor with its multi-agent mode, and GitHub with its “Agent HQ.” Cursor is increasingly a direct rival…

Read more

Today, we cover:

1. Amazon layoffs: AI or economy to blame? Amazon is doing more mass layoffs, claiming it wants to be more nimble. But are job losses really about US economic fears, and how Amazon’s retail business will be affected?

2. Industry pulse. Meta lays off 600 people from its AI division while hiring for another AI unit, $100K H1B fee will not apply to new grads, more AWS and Azure cloud outages, DEI costs the Python Software Foundation $1.5M, Citi blocks premium customers for weeks, and more.

3. OpenAI becomes for-profit, and takes on Chrome. Years overdue, OpenAI has finally changed its org structure to the for-profit model it’s been operating as. It has also launched Atlas, an AI browser meant to take on Chrome.

4. NVIDIA the biggest in Big Tech by a distance. NVIDIA is worth as much as Google and Meta combined, and is starting to expand from “just” selling GPUs to direct relationships with large companies such as Palantir, Eli Lilly, and Nokia.

1. Amazon layoffs: AI or economy to blame?

Online retail giant Amazon unexpectedly announced 14,000 job cuts earlier this week. The massive round of layoffs at the company follows other mass redundancies in recent years:

Read more