Rem Koning

@abhishekn @soumitrashukla9 This is awesome!!! Just got an R&R with @solenedelecourt and totally going to try this.

a recent example of how the practice of research changes dramatically with LLMs. the basic realization is that coding agents are great at webpages. so I've been turning my work into webpages. As someone who struggles to focus, this really helps me split tasks, and get started with long tedious work that I would normally avoid. Best example? Revise and response letters. i.e. responding to referees enumerating all the changes in a paper and providing a point-by-point response. I now have codex/claude spin up a custom website with each comment split out on a separate page, a gamified "completeness" meter, encouragements/affirmations along the way, and a text box for me to dictate responses that the system cleans up. there are also suggested points drawn from the revision, my own notes and its own thinking that helps me crafts the response on the side as a helpful resource. the system then grades responses, and gives me feedback on where I could do better, or the extent to which my response is incomplete/missing. it then takes all of my responses and creates a final latex response letter. the process of writing a response letter becomes, dare I say, enjoyable? highly recommend!

We've seen 40M projects built on Lovable and learned tons about what it takes to make a safe agentic coding tool. Lovable will be the first in our category to get the only certification covering the risks of agentic coding, by AIUC (@aiunderwriting). Read more about it and what we've done so far: lovable.dev/blog

Introducing the Built for Moms contest, in partnership with our friends at @WisprFlow, to celebrate everyone who builds for the moms in their lives. How to enter: 1. Build with Lovable and Wispr Flow 2. Submit: builtformoms.lovable.app 3. Vote on your favorites

New post on the difference between 3 notions of productivity gain from AI (AKA uplift). Uplift on old tasks (AI-speedup on tasks you do in avg 2022 day) Uplift on new tasks (AI-speedup on tasks you do in avg 2026 day) Uplift in value (AI increasing your goals accomplished)

@alexolegimas @GoogleDeepMind @ShaneLegg Amazing stuff!!! Excited to see what you cook up.

Some news: This week I am starting at @GoogleDeepMind as Director of AGI Economics on @shanelegg’s team. I will be joining the other amazing cross-disciplinary scientists researching AGI there. My team will study how frontier AI could reshape the economy: what happens to work and labor, how wealth and power are distributed, how institutions adapt, how AI agents shape markets, and what kinds of models can help us reason clearly about futures that may look very different from the past. I’m incredibly excited to help build this research agenda. If AGI changes how society operates, economics is going to be critical for shaping our shared future. Many more announcements soon.

@emmabwiles @johnjhorton Great stuff!

Excited to share a new HBR article on some of my recent research with colleagues at BCG about the phenomenon of AI Employees, and why governance is so important for integrating AI agents into your organizations work processes. hbr.org/2026/05/resear…

@melodykoh @robgo Love that summary @melodykoh and it's exactly what we tested. In our experiment treated group engaged in case studies, workshops, and peer groups to actually redesign how their startups operate.

Very interesting but not surprising. I think the key is HOW AI is implemented and incorporated into how you operate. Those who can actually redesign how they work probably saw the step-function gains, vs those who "used AI" without giving it context and the appropriate user authorship. I talked a bit about how this could play out inside companies as they think through what's possible: melodykoh.substack.com/p/the-build-vs…

My Partner @melodykoh has been on fire with her writing. But this is one of her best and most approachable recent posts. In AI, which is rising faster, the ceiling or the floor? I honestly don't know, but this is an amazing framing to the question. melodykoh.substack.com/p/in-the-meant…

@robgo @melodykoh Thanks @robgo! Next time you are on campus let's grab coffee.

@orgRem @melodykoh Very interesting. Thank you for sharing!

@thisisgrantlee and @antonosika I know culture came up in our HBS cases on @GammaApp and @Lovable but we didn't include an official code/doc on your cultures. Anything you can share with @bhalligan ?

One thing that really worked for HubSpot in the early days was that we published our "culture code" which, at the time, was quite unique. I'm waiting for someone to create a version of that for a truly ai native company. If you see a great one, send it along.

@AndreyFradkin @pawtrammell Would appreciate it!

@orgRem cc @pawtrammell per our recent discussion. I think it would be useful to clarify your definition of RSI and why it's different than the one implied here.

Are there other clean examples of recursive self improvement in technology? E.g., I am assuming some of Moore's law is that better computers helped us build better computers. Anything else?

NEW: I’m so excited to announce the Open Source for Science Fund, with $20M in anchor commitments from @biohub and @wellcometrust. Given the fast progress on AI, building this connective tooling could not be more timely or important. First RFP goes live in a week.

@mattbeane Ahahah! Turns out AI is just technology and always has been!

@orgRem This is true for literally all technology. Try that on for size.

@johnjhorton That's a good one. Thanks!

@orgRem i'd venture that better machine tools allowed the creation of better machine tools

@daveholtz fastcompany.com/91532226/andon…

@orgRem you can’t rule out the reverse causality story that the AI is running the cafes

In San Francisco for a few days mapping the AI frontier and honestly I think half of what makes SF amazing is great cafes. Explains why Stockholm is also cooking when it comes to AI. Suggests policy makers should probably just subsidize cafes to drive growth.

@griambau Let's just say most people are doing email, scrolling here, and listening to a talk so swapping out your phone for vibe coding seems like am improvement 😅 More realistically I would change format to 30 minute presentation and 60 minute vibe coding session to avoid the overload.

In a universe where people can listen to the speaker, think carefully, ask questions, code, correct code, check results, think about results *all at the same time*, this is a wonderful idea. In the world I inhabit, I can't see how this could make any talk pleasant or productive.

@GonHuertas Great question! Depends a bit, but I think with 2 hours (with the right pre-work) an incredible amount could get done. Vibe coding the IMF sounds quite fun!

@orgRem Interesting idea! Q: How would you envisage implementing this in internal presentations of applied research in, say, large international organizations? Same general approach, or do we need longer meetings? #AskingForAFriend

We should be vibe coding in seminars/meetings way more often. Instead of the speaker getting ~20 questions they implement a month later (or have an RA impement!), a seminar could be 90 minutes of analysis from everyone in the room. By end of day paper is 10x better and science goes faster. More on what I tried and why we need to start reinventing how we organize science to ensure AI leads to more *and* better science: thebindingconstraint.substack.com/p/why-only-ask…