☕️ Samuel Griff

It's a known limitation that an LLM is not able to catch its own mistake, similar to how humans are blind to their own mistakes until they learn. So unless Anthropic has trained a model specifically as a critic and cracked the code to help LLMs think in abstract terms (spoiler: they didn't), this is just marketing at its purest form as the link gives it away. You don't even have to open the link. The writing is on the wall with the vanity metric that's used to sell the snake oil: "share of PRs getting substantive review comments". 😅 Dude! LLMs can produce words... substantially! Not to belittle the achievements of Claude. It's a very useful tool. But I'm constantly baffled at the desperation of Anthropic and OpenAI to frame these LLMs as something larger and more impactful than they actually are! Last week we had their CEO saying "we're not sure if it's conscious" and mass media just put it on speakerphone completely missing these points: 1. We don't even have a definition of consciousness that's agreed upon 2. Blake Lemoine was the first to go viral with this claim 3. This kind of claim gets attention and that's exactly what they need in the absence of real progress (LLMs still hallucinate, context window is still limited, poor training data still embeds human malice in model behavior, etc.) 4. And this is important: to the average CEO who has no engineering and system thinking, this only spells one thing and we all know what it is. Try that AI PR review thing. But make sure to measure the right thing: 1. Do less people ship higher quality code? 2. What's the signal/noise ratio of AI generated PR comments? 3. Does involving AI in code review improve human understanding or hurts it? 4. When sh*t inevitably hits the fan, how does the AI involvement in creating the code hurt the TTR (time to resolution)? 5. How does AI overdose blind us to logical fallacies and errors? Look I'm not trying to hurt Anthropic here but I sure as hell care about the reliability of the services I pay for and my lifestyle depends on. I DO NOT accept subpar product quality just because engineers were forced or lured to overuse AI. Pls spread this message in your circles because our resilience as a society depends on mindful and deliberate use of AI and knowing when NOT to use it.

15

While using 𝗔𝗶𝗿𝗯𝗻𝗯 recently I noticed something small but very convenient. When a new message arrives from a host, it lands directly in your 𝘦𝘮𝘢𝘪𝘭 and you can simply hit reply to respond. No need to open the app. No separate inbox to manage. It feels simple, but it made me wonder: 𝗪𝗵𝘆 𝗱𝗼𝗻’𝘁 𝗺𝗼𝗿𝗲 𝗽𝗹𝗮𝘁𝗳𝗼𝗿𝗺𝘀 𝘄𝗼𝗿𝗸 𝘁𝗵𝗶𝘀 𝘄𝗮𝘆? After looking into it, the answer became clear: it’s not that simple behind the scenes. To make this work, platforms usually need 𝗲𝗺𝗮𝗶𝗹 𝗿𝗲𝗹𝗮𝘆 𝗶𝗻𝗳𝗿𝗮𝘀𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲, which involves:  • generating masked email aliases  • routing messages between participants  • parsing inbound email replies  • keeping conversations threaded correctly For large companies like Airbnb, it makes sense to build and maintain this internally. But for smaller teams, implementing and maintaining that infrastructure can be surprisingly involved. That got me curious, so I built a small API called 𝗥𝗲𝗹𝗮𝘆𝗖𝗼𝗿𝗲. The idea is simple: A developer creates a thread via API → RelayCore generates a masked email alias → participants communicate via normal email while identities remain private. No chat UI. No separate inbox. Just infrastructure for 𝗿𝗲𝗽𝗹𝘆-𝗯𝘆-𝗲𝗺𝗮𝗶𝗹 𝗺𝗲𝘀𝘀𝗮𝗴𝗶𝗻𝗴. I'm curious how others have approached this. Have you used platforms where you could simply reply to an email to continue a conversation (like Airbnb does)? Did you find it useful? And if you’ve worked on marketplaces or peer-to-peer platforms: • Did your team implement a third-party solution?  • Or built something like this internally? Would love to hear how you solved it. One thing that surprised me while exploring this was how complex inbound email parsing actually is. Email clients all format replies differently. https://www.relaycore.io #airbnb #marketplaces #devtools #startups

22

2 Comments

As LLMs evolve from static models into dynamic components of real-world applications, they must do more than perform tasks; they must also adapt to context, personalize interactions, and continuously improve through feedback. This blog outlines a practical, modular approach to designing adaptive, production-ready LLM systems using: - Supervised Fine-Tuning (SFT) for core task competence - Reinforcement Fine-Tuning (RFT) for behavioral alignment - Context-aware routing for persona and adapter control - RAG for real-time knowledge injection - Telemetry loops for continuous learning and refinement It also includes a curated toolchain (LoRA, trl, vLLM, Langfuse, and more). https://lnkd.in/gBtjeg_X

17

2 Comments

Spoke with an engineer in a 1:1 today. He shared an idea to improve our team's velocity. He proposed that since we had multiple services we should split the team for each microservice to make people more productive. I explained to him we don't have microservices but we do have a micromanager so there is no need to split the team. #leadership 🤗

14

2 Comments

Four patterns of AI native development: 1. From producer to manager 2. From implementation to Intent 3. From delivery to discovery 4. From content to knowledge Interesting patterns that seem legit based on my experience. Can’t vouche for the rest of the article/talk. I do agree that AI paradoxically forces teams to adopt good engineering practices (modular codebases, up-to-date docs, consistent style) they should have had all along, in order to get useful output from agentic tooling. https://lnkd.in/gGit4zAd

7

Interview tip: Ask your interviewer to show you a recent pull request. It reveals code complexity, review culture, what engineers care about, and how much is automated (CI/CD, checks, reviews). Ask about deployment and release frequency too - these details say a lot about what it’s really like to work there.

25

1 Comment

At Paddle, we're setting a new standard for developer-first infrastructure. Our API-first transformation with Paddle Billing has delivered significant wins, and our partnership with Postman is a testament to our engineering excellence. Here's a look at the outcomes: - 50% faster onboarding: We've reduced the time to first call by half. - 4x growth in engagement: Our API discoverability has quadrupled. - Shortened sales cycles: An improved developer experience is accelerating sales. Our API-first approach has also greatly improved internal collaboration, with shared workspaces and clear documentation providing instant context for our teams. Read the full Postman case study here: https://lnkd.in/g3K4bvHu

37

The engineers who'll thrive aren't the ones who write the best code. They're the ones who can articulate exactly what needs to be built, why, and how to verify it works.

8

1 Comment