Nhat Nguyen

Usually the early hires vibe well with the founders, so if the founders are jerks they will hire jerks or otherwise?

Wow! This is the incredible level of trust from partner, either business or romance.

@KSimback and the vector index everyone just calls the knowledge base

This is a good thread, but we need to correct the semantics - ontology, knowledge bases, and company brains are NOT the same thing They are closely related and often bleed into each other in practice so I understand why they can get confused > ontology is a formal semantic model of the concepts, entities, attributes, relationships, and rules in a domain It doesn't contain the actual data, think of it as a rule book for what kinds of things exist here and how they are allowed to connect An ontology without data or a knowledge base is useless, so they can't be the same thing > a knowledge base is the actual content repository It can be structured (e.g., a knowledge graph built on top of an ontology), semi-structured, or just raw unstructured documents + embeddings A KB can be useful on it's own because you can search it or run retrieval-based methods on it to get answers, but without an ontology you might struggle to understand how things in the KB relate or follow company rules > a company brain is the actual system on top that people and agents interact with It's usually an LLM + retrieval layer (often combined with orchestration, memory, tools, etc.) that makes the KB useful for answering questions, surfacing context, enforcing processes, etc. In the Cerebras write-up, what they built looks like a very well-engineered retrieval-focused knowledge base, and hats off to them for sharing it! It's an excellent practical 'company brain' implementation, but it doesn't appear to rely on a formal ontology, so again these are not all the same thing That's not a criticism, just pointing out the semantics I think the best modern systems will increasingly be hybrids - they have an ontology, a well-engineered knowledge base, and company brain They're just 3 layers in the stack to getting organizational truth into a place where you can reliably build agentic automation on top

@doublenickk grounding is the whole game here. a login screen at least fails loud, but an agent guessing your business rule fails quiet and keeps going.

STOP WASTING MILLIONS EXPECTING YOUR AI AGENTS TO MAGICALLY FIGURE OUT COMPLEX BUSINESS LOGIC ON THEIR OWN The biggest mistake developers make is treating an LLM like an omnipotent human worker. If you let an ungrounded agent loose on the web, a single unexpected login screen or security wall will completely break its loop, causing it to freeze, fail, or hallucinate a false success. The secret to building enterprise-grade autonomy isn't a better model, it’s implementing a strict Agent Harness. Think of it as the ultimate reality anchor for your AI: > Deterministic Guardrails: The harness operates entirely separate from the LLM, wrapped around the model to monitor its execution environment and intercept failures before they happen. > Smart Interventions: When the agent hits a roadblock (like a login window), the harness freezes the loop, securely injects the necessary credentials from the outside, and smoothly hand control back. > Controlled Iterations: It sets hard caps on maximum loops and handles rigorous final verification, ensuring the non-deterministic black box behaves exactly as you programmed it to. If you are still optimizing the context window with RAG and wondering why your production agents are failing, you're looking in the wrong place.

@coreyhainesco the linking is what makes a second brain actually queryable, good to see this compiles raw notes into a real wiki, not just a flat searchable pile.

I made a skill that gives my AI agent a real memory /second-brain — capture anything into raw/, compile it into an interlinked wiki, then query your own accumulated knowledge. It answers from what YOU'VE collected and cites its sources. I've been studying top YouTube videos this way — watch my brain answer "what makes a killer intro?" Part of Maker Skills — 18 free, open-source skills for founders & operators. Install in Claude Code: /plugin marketplace add coreyhaines31/makerskills

@rox_ai 91.3% at 1.03 cents a query is a genuinely hard spot to hit. the per-query cost is the number i'd watch as real traffic scales.

Introducing ask-web: Rox’s in-house web search agent. ask-web sits on the cost-per-accuracy pareto frontier of the hyper-parameter grid when compared to frontier labs and commercial search agent providers. The agent delivers 91.3% accuracy at 1.03 cents per query on real production prompts. It has been running in production for more than 6 months with continuous evals. Inference partners: @togethercompute, @baseten, @modal Commercial Search vendors benchmarked: @perplexity_ai, @ExaAILabs, @p0. Frontier Search vendors benchmarked: @OpenAI, @AnthropicAI Exa, OpenAI and Anthropic excel on accuracy. Parallel and Perplexity are cost-efficient. Here’s the breakdown:

@ArifromJelly @JellyjellyApp @iqramband @jellyvideochats more showing up than had tickets is the good sign. the walk-in energy is the part that makes these actually stick.

Our second ever founder/builder coworking session was a hit! 30+ people got tickets online and more showed up in person. Thanks to everyone who came by! Shout out to our sponsor @jellyjellyapp #Startup #Entrepreneur #Founder #NYC #Networking

@Voxyz_ai the backtest works because your past sessions are the honest record of how you actually work, not how you think you do. that gap is where most of my skills quietly rot.

Codex tip: take a skill or automation workflow you already wrote, throw it back at Codex, and have it run a 𝗯𝗮𝗰𝗸𝘁𝗲𝘀𝘁 against your past sessions. it rewrites the skill into how you actually work. 𝗰𝗼𝗽𝘆 𝗽𝗮𝘀𝘁𝗲 𝗽𝗿𝗼𝗺𝗽𝘁 𝗮𝘁 𝘁𝗵𝗲 𝗯𝗼𝘁𝘁𝗼𝗺. ・once a skill is done, have Codex evaluate it against the relevant sessions ・it finds the steps you always skip, the work you keep patching manually, and the order you actually run things in ・rerun this every few weeks and the skill keeps up with how your work changes the call behind it: skills should grow out of real work history. what you write from memory is the flowchart. sessions record how you actually did it and where the rework happened. session logs used to be leftovers from getting work done. now they double as an 𝗲𝘃𝗮𝗹 𝘀𝗲𝘁 for your skills. the way people write skills is changing. more builders are letting the agent read its own session history and reshape its own tools around it. 𝗽𝗿𝗼𝗺𝗽𝘁: read [skill name, workflow file, or folder path] and run a backtest against my past sessions related to it. prefer sessions from the same project and same task type. first list the samples you plan to use and why you picked them. if you can't access enough sessions, say so directly instead of guessing. compare what the skill assumes vs how i actually work. focus on: - steps i often skip, rewrite, or run more than once - work i still do manually that the skill never covers - the order i actually do things in - spots that keep causing rework, getting stuck, or needing extra clarification separate stable patterns from one-off exceptions. never rewrite a rule over a single anomaly. output a dry run first, nothing else: - keep / modify / add / remove - session evidence for each item - proposed diff and reasoning do not modify any files until i confirm. after i confirm, update the skill, run existing checks, and summarize before / after. if there's no existing way to verify, say so. finally, based on how fast new sessions accumulate, suggest a weekly or biweekly review. show me the schedule, scope, and trigger conditions first, then create the recurring task after i confirm. every scheduled review outputs a dry run only. no auto-modifying the skill without confirmation.

@_RohanShah_ @AaronLCannon Lol

@nhatng01 @AaronLCannon Umiami. So probably a lot of things

Why don't you check your LinkedIn inbox? My LinkedIn inbox:

@oliviscusAI the doc comes back formatted and done looking, so nobody rereads it, which is exactly when a wrong number ships

MICROSOFT JUST LOST ITS BIGGEST ADVANTAGE. someone built a single binary that gives ai agents full control over word, excel, and powerpoint. no office installation needed. it's called officecli. instead of guessing from raw xml, your agent renders the file to html or png and sees exactly what it built. → path-based addressing, no xml wrestling → render to html or png to catch layout bugs instantly → auto-detects claude code, cursor, and copilot no office license. no installation. no guessing. 100% free.

@ridark_eth rate limits hitting mid task are the villain here, ate a whole afternoon of mine once

The fact that this works is kind of breaking my brain.. you pay $20 to $50 a month per provider for Claude, GPT and Cursor, and still hit rate limits mid-task. this routes around all of them and taps ~1.6 billion free tokens a month for $0. somewhere Sam Altman is watching devs stack 90 free tiers into one endpoint so they never pay for a rate limit again... it's OmniRoute, a free open-source AI gateway (MIT). point every coding tool at one local endpoint and it handles the rest: > plug in any agent, Claude Code, Cursor, Copilot, Codex, Antigravity, through one config > 4-tier auto-fallback: your subscription → API key → cheap model → free tier, in milliseconds, so you never hit a wall > aggregates ~1.6B free tokens a month from 90+ providers' own free tiers, 11 of them free forever > compression cuts 15-95% of tokens off every request, stretching all of it further > runs locally and private, routing to the cheapest model that actually works the tools stayed the same. the bill and the rate limits just disappeared. Bookmark this before your next "you've hit your limit" message.

@0xfuckpoverty 'i just edit the result.' - the editing is judgment, and that stays human even when the model writes every line. i see the same in hiring, ai clears the busywork so people focus on the calls only they can make.

The CEO of Anthropic just dropped a massive reality check on the tech industry. 🚨 "My senior engineers tell me: 'I don't write code anymore. I just give the task to Claude Opus and edit the result.' Entire professions we've built for decades could simply disappear." If the people building the world's most advanced AI have already stopped manual coding, why are 99% of developers still doing it? The future of software engineering isn't writing syntax. It's orchestrating agents, defining architecture, and verifying outputs. We are rapidly shifting from creators to editors. Full breakdown of his Davos speech and what it means for the future of the job market below. Make sure to bookmark this! 👇

@contextconor @hyperspell @skyler_chan_ 😂😂

@nhatng01 @hyperspell @skyler_chan_ ✍️✍️ gotta add those to the roadmap, thanks for the ideas!

company brains are just the beginning for @hyperspell city brains next then country brains then world brains one day we'll partner with @skyler_chan_ to build a galaxy brain

@paularambles I looove the park. Surprised to hear that everyone hates it. Maybe I am not in sf long enough.

everyone hates on salesforce park but it’s literally the “society if” meme

@pmddomingos Aliens?

AI is eating software. What will eat AI?

@handotdev What if they smiled for a second for the picture and then stopped? They still needed to make a picture for social media which may bring in new customers :-)

Immediate disciplinary action incoming

@geertjanthomas @allenholub Indeed! Build and iterate.

@nhatng01 @allenholub That's why you need to build and ship all the time. The age of software in shrink wrapped boxes is behind us

I've been reading posts saying AI puts an end to "Agile." That's nonsense. AI agents are an irrelevance in an Agile context. Agents cannot tell you what to build, which is the key problem. It was never about coding, so much as writing the correct code. The point of Agile is to build software that people actually want/need through continuous learning. We release small increments to users/customers, get their feedback, and adjust. It is about having the flexibility to improve both what we build and how we build it as we learn. Historically, it addressed the problem of big up-front plans, which were the pre-Agile norm, not working, at least if you define "working" as building stuff people actually want to use. All that other garbage people cite (standups, sprints, &c.) has nothing whatever to do with Agile. The one exception is the retro, which is mentioned explicitly in the Agile Manifesto and is central to the learning process. You have read the Manifesto, haven't you? Doesn't seem like many of the "Agile" haters have. They seem to think that Agile is a process or framework. It isn't.

@dickiebush As long as there is life, there is hope.

My life got a lot better when I started viewing everything as rigged in my favor. Tragedy. Heartbreak. Victory. Risk. Devastation. Euphoria. Uncertainty. No matter what happened, I operated with a foundational belief that I would be fine either way. And magically, I was, once I started believe so.

@ash__lynns You can do both. Interviewing for the new job while staying at the current one. That is actually the way most people do it.

no greater pain than wanting a new job but having to stay with ur current one because the job market is absolute dog sh*t rn