Bart T. S.

Building a product for the 1st time has been a 🎢 , but here’s 4 key Aria updates: ↪️ Domain secured & waitlist page soon to be launched ↪️ We’re adding another staff/principal FS engineer to the team (DM me if you want to join an engineering 1st culture, cause GTM without an excellent product is nada) ↪️ We’re building the product alongside our pilot customers & appreciate everyone’s support here (DM if you want to 6.5x your recruitment productivity) ↪️ Have spoken to VC’s & seed investors these past months, and I’m open to this pathway (DM me if you’d like to learn about investing with us). We’re bootstrapping till now. Btw, I’m all in: nearing the end of cash reserves, selling stocks next (even my Nvidia 🫣), selling 🏠 after if needed (homelessness here I come haha). Beyond excited, Abdullah

15

4 Comments

I am working on the updated ID translation for "Keep A Changelog". This project represents something I care about, and I'd like to share it. What is "Keep a Changelog"? Keep a Changelog is an effort, built around the idea that those who use or work on software deserves an easy-to-read, complete and unambiguous list of changes that happens to it. The goal? A better convention for how we keep and write a changelog. Visit the site and give it a read: https://lnkd.in/gqEH8sBH If you find yourself agreeing with these principles, you can help me in translating it! https://lnkd.in/gEkamsQV

2

🇯🇵 Agentic Challenge. Write an agent in the least code possible. Here’s an agent in a single line using Deno / Typescript & Japanese kanji Agent.ts const ai = new (await import("ruv.io")).default();Deno.serve(async r=>Response.json((await ai.chat.completions.create({model:"o1-mini",messages:[{role:"system",content:"英語で返信。簡潔に"},{role:"user",content:(await r.json()).p}]})).choices[0].message)) Run it. OPENAI_API_KEY=sk-xxx deno run --allow-net --allow-env agent.ts Why am I using Japanese for guidance? Using Japanese minimizes characters because kanji symbols can convey complex instructions succinctly. A single kanji character often represents entire concepts that would require multiple English words, reducing the overall length. Additionally, Japanese syntax tends to be more compact, allowing concise guidance without sacrificing clarity. This efficiency is ideal for keeping the agent’s code minimal while still providing necessary instructions, ensuring functionality with fewer characters and optimizing the agent’s performance.

88

36 Comments

So here’s the long and short of it: Rabbit R1 is a $200 piece of hardware (which happens to look cool, because Teenage Engineering is a great design studio) that says “ET phone home” for every action you take. Under the hood, it appears to be a bunch of Playwright scripts. (With the exception of “Play a random song.” This is hard coded to be a Beatles song. Which I find hilarious.) Playwright is dope. You should learn how it works. I’m not throwing shade for using tools that are dope if you’re making a dope product. I just don’t understand how they got like $300M in venture funding to make… a voice-to-text NLP processor that runs a bunch of Playwright scripts and then calls them “apps” or whatever. (A relevant limitation here is going to be the web interface, because if even one button moves, your app breaks. Hence why all the Rabbit apps constantly break.) What did we learn here today? Playwright is useful. Skip the $200 hardware and latency. https://playwright.dev/

29

5 Comments

Finding PMF is easy. Yet many companies die trying to find it. But if you can sell finding PMF is easy. Most people that buy technology don't know they have a problem. It's up to the founder to uncover pain and create the need. But the problem is many technical founders can't sell. So they focus on the technology. And opportunities don't covert. This has nothing to do with PMF. It's a shame because many great products get killed because they run out of runway and get they get put in the no PMF bucket. Learning the basics makes such a difference. I've set out on a mission to help as many founders I can sell better. It's why I'm really excited about our founder community Cosmic Collective. So if you are a founder of VC backed technology company and want to know how to sell, let me know. I'll show you how to find PMF!

11

This is disruptive

12

【LM Studio Report on M4 MacBook Pro】 Tried latest 70B Llama on my M4 Mac 💻 (You have to set off the default restriction by “Model Loading Guardrails” if Mac's memory is small), It did not freeze but output was so slow… 2.8 ~ 3.0 tok/sec While 7~8B LLMs are able to write 30 ~ 38 tok/sec ✏️ I guess my Mac can perform alright up to around 30B. My Mac spec is as follows. BTW, wanted buy a Max chip but could not afford this time… Chip: Apple M4 Pro CPU Cores: 14 (10 performance and 4 efficiency) GPU Cores: 20 Memory: 48 GB One more thing, as far as what I have asked, I don’t think Bigger LLM answers are always remarkable comparing to smaller ones. 

3

Interoperability in / of Data Clean Rooms is an often confusing topic! Why? Well on one level, clean rooms by design are not actually supposed to be interoperable with one another. Odd statement coming from someone who has talked about the concept of clean room interoperability? Not at all. This is because the interpretation of interoperability is broader than the notion of trying to make RMN A's clean room talk to RMN B's, or for AMC to link to ADH. Each are distinct islands of data, and that's not just okay it's actually part of the value prop. All this said I actually DO see big opportunities for and areas of interoperability in a broader sense within the technology, so here's three thought-starters... 🚪 Access & Productivity: Each DCR, whether it's a custom solution or something "off the shelf" (from i.e. a walled garden) typically has its own front end, maybe a specific taxonomy, privacy rules & PETs, unique data spine, use cases... Whilst these DCRs are not equal under the hood, the emergence of APIs & cloud primitives allows us to execute commands into warehouses, inject queries into the clean rooms, and to pull aggregated tables back. This represents a big opportunity for software to act as a unified access layer. This form of interoperability is going to be mission critical for larger brands who are now headed into a world of 00s of clean rooms. Users will not have the capacity to operationalise at scale without help. ⚙ Data Integration: For clean rooms to be at their most useful they require data from multiple sides. This might be as simple as a publisher providing their ad logs + CRM on one side, and a marketer conversion data on the other. Regardless of the data recipe, today it is necessary for clean room providers to high levels of interoperability with the existing sources of business & customer data. A clean room that can only read data from it's own storage layer / bunker / enclave / vault (you can insert any secure-sounding word here as a replacement for S3 bucket) is the antithesis of interoperability, restricting the ability to work with partners outside of the ecosystem. Think closed data networks vs open data networks. I don't have enough characters to go into the AI opportunity linked to this one here!! 📦 Standardisation: A challenge related to point 1 is that in the first wave of clean room use cases, everything has been custom. This doesn't scale well and makes it hard, for example, to compare outcomes across clean rooms even on an aggregate level. In 2025 we're going to see a big rise in agencies, brands + pubs all deploying deliberately more consistent & universal analyses across clean rooms. Doing so will make it easier to compare the velocity and shape of data across different clean rooms, thanks to i.e. common dimensions, even if we can't necessarily look across direct row level matches for privacy reasons. #datacollaboration #datacleanrooms #adtech #martech

73

20 Comments

I've known Tobi for over twenty years now. Right from the earliest days of Ruby on Rails, when he was building Snowdevil, which eventually became Shopify, to sell snowboards online. His first commit to Rails was from 2004, which improved the ergonomics of controller testing. Just one out of the 131 commits he made to the framework from 2004-2010 -- a record still good enough to be in the top 100 all-time contributors to Rails! But Tobi's contributions to Ruby on Rails extend far beyond his individual commits to the framework, creating Active Merchant and the Liquid templating system, or serving on the Rails Core Team back in the early days. With Shopify, Tobi more or less single-handedly killed the zombie argument that Rails couldn't scale by building the world's most popular hosted e-commerce platform and routing a sizable portion of all online sales through it. In the process, Tobi built an incredible technical organization to support this effort. Shopify employs a third of the Rails Core Team, developed the YJIT compiler for Ruby, and contributed in a billion other ways. They are without a doubt the most generous benefactor in the Ruby on Rails world. So when Tobi asked me whether I'd be interested in joining Shopify's board, I needed no pause to consider the invitation. OF COURSE I WOULD! But to be honest, it wasn't just a reflexive answer to service the gratitude I've felt toward Shopify for many years. It was also to satisfy a selfish curiosity to wrestle with problems at a scale that none of my own work has ever touched. Both in terms of the frontier programming problems inherent in dealing with a majestic monolith clocking in at five million lines of code, and the challenge of guiding thousands of programmers to productively extend it, Shopify deals with a scale several orders of magnitude beyond what I do day-to-day at 37signals. That's interesting! So too is the sheer magnitude of the impact Shopify is having on the world of commerce. While much of the web is decaying to enshitification and entropy, Shopify stores stand out by being faster to browse, quicker to checkout, and easier to trust. That's enabling a vast array of individual entrepreneurs and businesses to have a competitive shopping experience against the likes of Amazon, without needing huge teams to do it. It's always a delight when I find a cool store, and I learn that it's running on Shopify. As I spoke with Tobi about on the announcement show, this was really hammered home after I got into mechanical keyboards. Seemingly every single vendor of thocky and clicky keyboards use Shopify! And when I see that, not only am I sure that buying won't be a hassle, but I also know I'm not going to get scammed. That's the Shopify magic: Leveling the commercial playing field between some obscure keyboard maker and the consolidated titans of e-commerce. And now I get to help further that mission from the inside! What a treat. Thanks Tobi!

1,595

39 Comments

🍀 Intercom 🥰 "Castle of Code" (Wes Sheeran lots of content) D           G         Bm  A When I launched my app, I hit the wall D            G           Bm     A Too many questions came, couldn’t manage them all D          G          Bm       A But then I found Intercom, to handle the flow D           G  Bm         A And it scaled with me, made my support grow G      A          D Found my way through the clutter here G Saved time and made it crystal clear G      A         D         G Automations running day and night, and now I’ve grown G       A And I can’t wait to go home D      G  Bm     A I’m on my  way, automating tasks D       G   Bm     A Through those busy  days, answering fast D      G      Bm    A And I miss the way you make me feel, oh it’s real D        G   Bm    A         D Now I watch the tickets fade, over the castle of code D         G       Bm           A Fifteen open chats, no panic at all D           G           Bm        A Intercom handles it right, every ticket, big or small D        G          Bm      A Customer delight from the very first click D          G  Bm         A I knew I’d struck gold, Intercom does the trick G      A          D Now our team is on the rise G No more endless back-and-forth replies G      A         D       G Chatbots solving questions fast, we’ve reached new heights G      A And I can’t wait to go live D      G  Bm     A I’m on my  way, automating tasks D       G   Bm     A Through those busy days, answering fast D      G      Bm    A And I miss the way you make me feel, oh it’s real D        G   Bm    A Now I watch the tickets fade, over the castles of code D    G  Bm    A Over the castles of code D    G  Bm    A Over the castles of code Bm G D A Bm       G          D      A One friend’s in marketing now, one’s on DevOps patrol Bm       G          D    A One’s automating chat flows, one’s on support roll Bm     G            D      A But all my team knows Intercom’s gold, it’s made our dreams unfold Bm       G         D    A These bots raised me and I-I-I, can’t wait to go home D      G  Bm    A And I’m on my  way, I still remember D      G     Bm     A Those old support days when we struggled together D      G      Bm    A And I miss the way you make me feel, oh it’s real D        G   Bm    A Now I watch the tickets fade, over the castle of code D    G  Bm    A Over the castle of code D    G  Bm    A         D Over the castle of code CC Des Traynor 🤓

2

Cost optimisation regardless of startup or not should always be part of hygiene and routine than a one-time activity in business. This helps to avoid crisis situations during tough times 🙌

1

There is a $7.5B company that: • Started with $0 funding • Has 100M+ active users • Generates $1B+ in ARR And now powering the open-source revolution. Here’s the story of GitHub: Tom and Chris were both working at CNET when they met in the mid-2000s. They were both passionate about software development and knew that developers needed better tools to manage and share code. In 2008, they started working on a side project that would change their lives. Git was a version control system at the time. But it had terrible UI and no functionality for collaboration. Their idea: build a web-based platform to host the Git repository. It would revolutionize the way developers collaborated on code. They launched GitHub in April 2008 with just a handful of users. Early on, they bootstrapped with an intense focus on customer experience and on exceptional UI. And they fully embraced the open-source community. The commitment paid off... Three years later, they had over a million developers on their platform. GitHub became the focal point of the open-source revolution. Hosting and collaborating on code was never so easy. Even big tech (FB, MSFT, Google etc) became customers. From there things skyrocketed 👇 By 2012, the biggest VCs started noticing. GitHub raised $100 million in Series A funding led by Andreessen Horowitz Their growth was explosive. By 2015, they had 9 million users and hosted over 21 million repositories. How did they scale so rapidly with a relatively small team? Simple answer: They leveraged talent all over the world. GitHub adopted a distributed team model early on, hiring talented developers from around the world. Scott Chacon, a co-founder, worked remotely from Mexico, setting the precedent for a global workforce. GitHub attracted top talent from regions like Eastern Europe etc. GitHub's distributed model also allowed for continuous development cycles with teams in different time zones. By 2018, GitHub had revolutionized the developer experience. And Microsoft bought them for $7.5 BILLION. All the co-founders are billionaires. Not bad for a side project. GitHub, WhatsApp, Wayfair and many giants were built on global talent. Trained pros take on rote execution at a fraction of the cost, while top teams focus on strategy and product. GrowthAssistant embeds elite, college-educated pros from the Philippines directly on your team. We find and place talent with the EXACT skillset you need. DoorDash, HubSpot, True Classic, and 200+ others use us Check it out:

23

2 Comments

When engineers say "no," it reveals flaws in your pitch or opportunity, helping refine your approach Here are a few common mistakes 👉 𝗖𝗼𝗻𝘁𝗮𝗰𝘁𝗶𝗻𝗴 𝗽𝗼𝘁𝗲𝗻𝘁𝗶𝗮𝗹 𝗰𝗮𝗻𝗱𝗶𝗱𝗮𝘁𝗲𝘀 - 1. Not including basics things like work location, comp range, and tech stack 2. Not highlighting what makes the role/company relevant for them 3. Targeting the wrong (or not enough) people 𝗙𝗶𝗿𝘀𝘁 𝗶𝗻𝘁𝗲𝗿𝘃𝗶𝗲𝘄𝘀 - 4. Jumping right into technical questions before creating buy-in 5. Not pitching your company's vision 6. Not understanding why the candidate is interviewing and what they're looking for in their next role 𝗟𝗮𝘁𝗲𝗿 𝘀𝘁𝗮𝗴𝗲𝘀 - 7. Assuming candidates are interested, without understanding their decision framework 8. Not debriefing and addressing concerns after every interview 9. Extending offers without pre-closing Good recruitment involves a lot more than just finding top talent Have a great weekend!

8

1 Comment

Wow! This looks BIG. Claude 3.5 released 12 June. Free, pro versions. "We're all developers now" example from Jo Zhu, Anthropic, via an examples thread on X/ Twitter from #MinChoi. Jo Zhu created a backgammon app in 4? mins, with 2 pics! - 1 of rules, 1 of setup. She used Artifacts - a set of capabilities available with Claude, "designed to enhance the functionality and versatility of their AI models." Started using the free version yesterday. Comparing responses to the paid OpenAI GPT4o - initially looking impressive. Time to get experimenting! I've been looking for a Go/Wéiqí/ Baduk game app ? :) Easy to get started. Requires registration https://claude.ai/ Anyone have the Claude 3.5 pro plan? Worth getting? Jo Zhu's X example thread https://lnkd.in/gNX5yKu8 Min Choi's X examples thread https://lnkd.in/gnTybvEe #AI #GenAI #Claude #Anthropic #LLM

6

1 Comment

Last week, and acquaintance reached out because she was concerned that a family member was being scammed, and she wanted to see if I could help. Hearing the situation, I agreed that it sounded like a crypto ponzi scheme, and this person was being asked to put in more and more money in order to get to the "next level" of commissions. And, of course, they couldn't get their money back out until they completed some elaborate course of investments and actions. I spoke with both of them, explained that it certainly sounds like a scam, and has all the hallmarks of a dodgy sales funnel. This person finally agreed that the money they put in was gone (something like $90k), and to step away. Here is the interesting bit - this person is a lawyer. They are highly educated with a ton of experience dealing with shady people and fraudsters... but the combination of the slow ramp up, ignorance about crypto, and the sunk cost fallacy combined to make an effective trap. We should all be sure to educate our families and friends about the dangers on ever more sophisticated online scams, and remind people that there is no true path to "easy money." Any true "get rich quick" schemes will be played out and milked dry long before the average person gets wind of it. We can't stop the scammers, but we can help educate our communities and make it harder for them to take advantage of those close to us. Personally, I tell all my friends and family to be super suspicious of "Get rich quick" schemes, and that they are welcome to ask for my help evaluating something. I may, or may not, know whats going on... but just taking a moment to break the hype cycle and explain the situation to a 3rd party can often be enough to reveal how suspicious a scam really is.

3

A big part of the reason that companies are going ga-ga over AI right now is the promise that it might materially lower their payroll for programmers. If a company currently needs 10 programmers to do a job, each have a cost of $200,000/year, then that's a $2m/year problem. If AI could even cut off 1/4 of that, they would have saved half a million! Cut double that, and it's a million. Efficiency gains add up quick on the bottom line when it comes to programmers! That's why I love Ruby! That's why I work on Rails! For twenty years, it's been clear to me that this is where the puck was going. Programmers continuing to become more expensive, computers continuing to become less so. Therefore, the smart bet was on making those programmers more productive EVEN AT THE EXPENSE OF THE COMPUTER! That's what so many programmers have a difficult time internalizing. They are in effect very expensive biological computing cores, and the real scarce resource. Silicon computing cores are far more plentiful, and their cost keeps going down. Let's return to that $200,000/year programmer. You can rent 1 AMD EPYC core from Hetzner for $55/year (they sell them in bulk, $220/month for a box of 48, so 220 x 48 / 12 = 55). That means the price of one biological core is the same as the price of 3663 silicon cores. Meaning that if you manage to make the bio core 10% more efficient, you will have saved the equivalent cost of 366 silicon cores. Make the bio core a quarter more efficient, and you'll have saved nearly ONE THOUSAND silicon cores! But many of these squishy, biological programming cores have a distinctly human sympathy for their silicon counterparts that overrides the math. They simply feel bad asking the silicon to do more work, if they could spend more of their own time to reduce the load by using less efficient for them / more efficient for silicon tools and techniques. And I actually respect that from an artsy, spiritual perspective! There is something beautifully wholesome about making computers do more with fewer resources. I still look oh-so-fondly back on the demo days of the Commodore 64 and Amiga. And that's the kind of work I've been doing for said twenty years! Making business software and selling it as SaaS. That's what an entire industry has been doing to tremendous profit. It's been a bull run for the ages, and it's been mostly driven by programmers working in high-level languages figuring out business logic and finding product-market fit. So whenever you hear a discussion about computing efficiency, you should always have the squishy, biological cores in mind. Most software around the world is priced on their inputs, not on the silicon it requires. Meaning even small incremental improvements to bio core productivity is worth large additional expenditures on silicon chips. And every year, the ratio grows greater in favor of the bio cores. https://lnkd.in/dRMKS4pT

1,114

95 Comments

Stacked Diffs is the best kept secret of Meta and we copied it 💥 Stacking has been months into working at HyperTest. But now this is our default way to build and ship code because it does not block our devs for reviews. Devs love it and also complain why did we not implement this earlier. But why did we need to move to this? The traditional development workflow is this: 1, Finish a feature 2, Submit a PR 3, Await PR approval 4, Merge back into main once approved Once the PR is merged, start work on the next feature. This process takes a long time! The biggest time waster is that PRs sit waiting for review for hours or days. This wait becomes longer if current PR is dependent on a feature in another PR still under review This where stacking comes to rescue Stacking lets devs make many small PRs easily, without waiting for a review. To give a practical example, it’s pretty common that once a dev builds a feature, he wants to build the next feature using the completed one. With stacking, you don’t have to wait for the first feature to be approved and merged; you just keep working! The idea behind stacked diffs is that you can keep working on your main branch, and worry about reviews later. This is why very high performance teams, like the ones Meta, Google, Amazon, Rubrik and others use stacking as the only default way to write and ship code. But stacking is much easier to get started with in smaller teams. The learning curve was negligible but returns are great! We love it. #softwareengineering #stacking #git

16

Feeling lost in the sea of no-code/ low-code tools? I’ve been there! After trying to find a single "magic tool" that could handle everything—from building versatile software to managing complex integrations—I realized there’s a better way. Here’s what worked for me: Instead of relying on one tool to do it all, I found that using ChatGPT as a guide made all the difference. My custom assistant now helps me map out each project step-by-step, directing me to the best AI, automation, or no-code tool for each specific task. ChatGPT even generates ready-to-use code for each step. One key insight: don't prompt. ChatGPT understands intent communicated through human language better than it understands scripts, so explaining your project as if you’re talking to a colleague yields better results. If you’re exploring no-code or low-code solutions, this approach could save you countless hours of trial and error. Happy to elaborate further if anyone needs help to get moving.

7