Michael Ryberg Hartmann

Curious about technology trends and fascinated by the ever growing platforms of the world

[Paweł Huryn @PawelHuryn]

https://t.co/rnkK6kImP0

[Boris Cherny @bcherny]

We've been working on this for a while -- it's impressive (and scary) to see the kinds of security issues it has identified.

Rolling out slowly, starting as a research preview for Team and Enterprise customers.

[Boris Cherny @bcherny]

I'm Boris and I created Claude Code. I wanted to quickly share a few tips for using Claude Code, sourced directly from the Claude Code team. The way the team uses Claude is different than how I use it. Remember: there is no one right way to use Claude Code -- everyones' setup is different. You should experiment to see what works for you!

[Amir Salihefendić @amix3k]

https://t.co/IBPGhtWJbI

[Aakash Gupta @aakashgupta]

Claude Code tasks are epic. Here’s your primer.

[CJ Hess @seejayhess]

Claude Code tasks might go down as the most impactful feature they have built. The Swarm is here.

[Vercel @vercel]

https://t.co/A0TmnEzUVa is an open ecosystem for finding and sharing agent skills.
Add a skill to any agent with:

▲ ~/ npx skills add <owner/repo>

[Kieran Klaassen @kieranklaassen]

compound-engineering v2.27.0 is out!

Big shoutout to our community contributors:
→ tmchow - interactive Q&A for plans + incremental commits
→ ashwin47 - style guide fixes
→ @rbouschery - docs updates

Your PRs make this plugin better for everyone 🙌

https://t.co/xmotbwmpXT

[Aakash Gupta @aakashgupta]

Ralph may be the most important thing to learn in AI right now.

Someone used this technique to deliver an entire contract that would have cost $50,000 to outsource. Total API costs: $297. At a Y Combinator hackathon, teams shipped 6 working repositories overnight using the same approach. Geoffrey Huntley, who invented this, built an entire programming language over 3 months while barely touching his keyboard.

Ralph is a bash loop that runs an AI coding agent repeatedly until a task is done. That’s it. While you sleep, while you eat dinner, while you do anything else, the loop keeps going. It picks up a task, builds it, checks if it works, saves the progress, picks the next task. When you wake up, features are finished.

Why this works when normal AI coding fails: most people open Cursor or Claude with a vague idea and no structure. 45 minutes later they’re fixing the same bug for the third time. The AI forgot what they were building. The context got polluted. They’re frustrated and nothing shipped.

The problem is task size. One feature has 20 pieces. The AI tries to hold all of them at once. It can’t. It hallucinates. It contradicts itself. You end up babysitting.

Ralph fixes this by breaking work into pieces small enough that the AI finishes each one before it forgets what it’s doing. Each loop iteration starts fresh. Clean context. No accumulated confusion. Memory persists through git commits, a progress file, and a task list. The AI reads what happened last time, learns from mistakes, picks the next task.

The workflow:

Step 1: Describe what you want in plain language. Talk for 2-3 minutes. “I want users to filter tasks by priority. High, medium, low. A dropdown with all three options plus ‘all’. Selecting one filters the list immediately.” Then tell the AI to convert your rambling into a formal requirements list.

Step 2: Break requirements into atomic tasks with binary success criteria. Good: “Add a priority column that defaults to medium.” Bad: “Make it work well.” The AI needs to know when it’s done without asking you. Pass or fail. Yes or no.

Step 3: Run the loop. Ralph grabs a task, builds it, runs tests, commits if it passes, moves to the next one. You set a limit (10 rounds, 14 rounds, 20 rounds). It runs until everything passes or hits your limit.

The math is brutal. A typical Ralph run: 10 iterations, roughly $30 in API costs. A senior developer costs $400-600 per day fully loaded. If Ralph gets you 90% there and you spend an hour on cleanup, you just converted an 8-hour workday into 1 hour plus $30.

That’s $500+ of labor arbitrage per feature. Compound that across a month of building.

Ralph can replace the majority of outsourcing for greenfield projects. The technique is “deterministically bad in an undeterministic world,” meaning when it fails, it fails predictably. You tune it like a guitar. When Ralph screws up a specific way, you add a guardrail to the prompt. The failures become instructions.

The real shift is in role. You stop being the person who writes code. You become the person who writes requirements, defines acceptance criteria, and reviews output. Product designer, not engineer.

“Software development as a profession is effectively dead. Software engineering is more alive than ever.” The people who understand systems, architecture, requirements, and quality will thrive. The people whose value was typing code faster are already obsolete.

The gap forming right now: builders who know Ralph are shipping 5-10x more than everyone around them. Nobody understands how. The technique is free and open source. The barrier is just knowing it exists and spending 30 minutes to understand the pattern.

Three months from now this will be in every YouTube tutorial and paid course. The window is right now, while most developers are still prompting Claude one function at a time and wondering why AI “doesn’t work for them.”

https://t.co/Z3WtUel1uH

[Nutshell Group @NutshellGroup]

"Technology enables organizations to use financial solutions according to their unique requirements." Michael Hartmann (@michartmann), Founding Partner, RH Consulting.

#TheFutureSummit #TheBigPicture

[Arab News Pakistan @arabnewspk]

#WATCH: UAE-based financial technology firm evaluates Pakistan for business expansion. “We’re on the mission of trying to kind of move the needle on financial inclusion,” @FintechGalaxy's @michartmann tells Arab News
-
https://t.co/Pxs4VgqDO0

[Chris Gledhill @cgledhill]

The Rise of Banking-as-a-Service

https://t.co/HhOvAg3Hzo #fintech #baas #banking

[XeroAPI @XeroAPI]

We welcome Profitdiagnostix to the @Xero App Marketplace today. They deliver specialised reports for service professionals like veterinarians and dentists. Get more info here: https://t.co/FIJ2uLWeHQ ^GBR

[Michael Ryberg Hartmann @michartmann]

Great points from Mr. Skinner. Couldn't agree more. Too many are throwing the buzz words around without truly understanding them.
Platform, marketplace, ecosystem ... buzzword bingo! - Chris Skinner's blog https://t.co/nccDijjvaz

[Michael Ryberg Hartmann @michartmann]

@runemai There is the off chance that people in banking have done other things before joining the corporate world ;)

[Michael Ryberg Hartmann @michartmann]

@gatsbyjs Damn you guys rock !

[Michael Ryberg Hartmann @michartmann]

@gatsbyjs Really love the way you guys have build your new doc v2 site - any chance you could share the underlying tech solution to be inspired from :)

[Jon Schäffer @JonSchaffer]

If you spent your summer holiday thinking... I can do better, go faster, grow bigger!!! Then join a new super-team at #danskebank to drive growth through #commercial and #fintech #partnerships ... #jobsatisfaction https://t.co/2DEcGx3Ihn

[Megan @MeganCaywood]

Great article by @FinancialBrand @JimMarous: “While much of the discussion about #OpenBanking and open #APIs is in the U.K., the impact of building a marketplace #platform in financial services should be part of every bank and credit union strategic planning process globally.” https://t.co/DeBR9b2f0F

[Atlassian @Atlassian]

The traditional notion of "climbing the ladder" by accumulating your years of service is obsolete. https://t.co/SxmiU0n8B4