Just do what you can with what you have where you are || Incremental work pays off : 1% daily gain → (1.01)^365 ≈ 38 → ~38x growth/year
Canada 🇨🇦
Joined December 2010
- Tweets419
- Following2.2K
- Followers54
- Likes300
ZX🇨🇦 retweeted
We're giving away a CONTROL Resonant custom wrapped NVIDIA GeForce RTX 5080 to celebrate its official release with #RTXON.
To enter:
🟢 Share this post
🟢 Comment #RTXON
T&Cs: nvda.ws/4z24Gcq
ZX🇨🇦 retweeted
🚀 System Design for the LLM Era
Patterns and Principles for Production-Grade AI Architecture
Building an LLM-powered demo is easy.
Building an AI system that is reliable, scalable, secure, cost-efficient, and production-ready is a completely different challenge.
This book explores the architecture behind modern AI applications, including:
• LLM application architecture
• RAG and vector search systems
• Agentic AI workflows
• Prompt and context management
• Model routing and orchestration
• Caching and latency optimization
• Evaluation and observability
• Guardrails and AI security
• Scalable inference systems
• Cost optimization
• Production monitoring
• Failure handling and reliability
amzn.to/46bOQje
If you’re building real-world applications with LLMs, RAG, AI agents, or generative AI, system design is becoming one of the most important skills to learn.
The next generation of software engineers won’t just design distributed systems—they’ll design intelligent systems.
ZX🇨🇦 retweeted
Over 300 original figures!
An Illustrated Guide to AI Agents is now OUT on Kindle and other Ebook stores! Presses are rolling, print copies ship in a few weeks.
@MaartenGr and I spent the last year and a half making the agent stack legible: memory, tools, planning, evaluation, multi-agent systems, and code agents. We're ecstatic to finally be able to share it with you!
Agents are being adopted faster than they're being understood. Our bet is that a small number of concepts will outlast everything else in this field, so we spent the time identifying them and explaining them properly.
Concepts and code. You build a working agent yourself along the way.
Order here:
Amazon: amazon.com/Illustrated-Guide…
O'Reilly: oreilly.com/library/view/an-…
ZX🇨🇦 retweeted
Trending repository of the month 🏆
archify
Agent skill for beautiful, verifiable architecture, workflow, sequence, data-flow, and lifecycle diagrams—self-contained HTML with motion and crisp export.
Last month: 29,539 ⭐
Total: 40,866 ⭐️
github.com/tt-a1i/archify
ZX🇨🇦 retweeted
i’ve been building and using agents since the gpt-4 days
back then, the model couldn’t even reliably edit files. it couldn’t generate a valid unified diff, couldn’t preserve white spaces in find/replace tool calls, plus various kinds of quirks
we implemented many harness level tricks to compensate, it got a lot better than using the raw model, but it still sucked
then sonnet 3.5 v2 came out and it was trained with a coding harness with RL. it knew how to find/replace reliably. it knew how to compose good bash commands to get what it needs. all the tricks we did were reverted
that was the first time i internalized the bitter lesson incompleteideas.net/IncIdeas…
most of what we are doing in the harness today, all the smart context engineering tricks, all the useful markdown files we throw into our repo, they will all go away
eventually, the models just know
there will be a model that will do incredibly efficient and smart compaction that preserves what the continued session actually needs
there will be a model that know how to plan with you better than any skill you can find today
there will be a model that will do perfect code reviews with just the right feedback
the models just know
i believe as humans today, most of us should not be fiddling too much with harness tricks. if you don’t believe they will go away, you can surely still believe they will keep changing rapidly to a point where anything you learn becomes obsolete every several weeks
instead, focus on what i call “the 3 fundamentals” -
1. understanding the world
this is your input. how does the world work? what are people doing? what problems do they have?
knowing how to ground yourself with real world knowledge helps you avoid hallucinations, and work on things that actually matter
2. first-principles thinking
this is your compute. with the real world understanding you gathered, what insights can you derive that’s likely to be true? what predictions can you make about the future?
being able to think in a disciplined way is what allows you to arrive at useful conclusions that can guide your actions
3. articulating our thoughts
this is your output, and it’s your AI’s input. it may sound easy but it’s a real skill. not everyone can communicate effectively, whether it’s to humans or AI
with AI eventually becoming incredibly capable, the clarity in the articulation of our intent is the main, if not the only, bottleneck
these 3 fundamentals do not shift as models improve. they only become more and more critical as the bitter lesson manifests
for the vast majority, i suggest spending your time on what will still matter in a year, five years, a decade. let the geeks play with what’s hot this week - in the end, you will see that you didn’t miss anything
ZX🇨🇦 retweeted
since i quit big tech a few months ago and started my own adventure, i have declined opportunities from anthropic, openai, google, microsoft, many other big companies, startups and investors
i respect most of these companies and don’t take their gesture lightly. but my choice to go solo was very, very deliberate
i did so because i strongly believe AI can now give every individual incredible leverage to achieve their ambitions that were previously impossible without a company. in many cases, an individual with extreme clarity and cohesion in their vision should actually be able to move faster than a company. and i’d like to help make that a reality for more people
but to truly understand how that would manifest, i must live and breathe that experience myself. i need to prove how far one person can go and figure out what the real bottlenecks are
that’s the driving force behind everything i did over the last couple of months and why i’m not for hire right now
ZX🇨🇦 retweeted
If you want to get into YC, don't focus on getting into YC. Focus on building stuff and understanding your users. That's what gets you into YC.
ZX🇨🇦 retweeted
(1.01)^365 ~ 38
"First do it, then do it right, then do it better."
Just start. The journey to success often begins with a single step, but that first step can be the hardest to take. It's easy to get caught up in the fear of failure or the desire for perfection, but I hope this quote I first shared in 2013 can be a reminder of the importance of simply getting started as we go into 2024.
Just Start Somewhere
"Start slow if you have to. Start small if you have to. Start privately if you have to. Just start." - James Clear
Taking that first step doesn't require perfection or immediate mastery. The key is to overcome inertia and take action, as this action will lead to progress, learning, and (if you’re lucky and consistent) ultimately success.
When you start, you allow yourself the opportunity to grow, adapt, and move forward.
The Power of Starting
Beginning a new project or habit often feels daunting. According to psychologists, we tend to overestimate the pain of starting and underestimate our ability to persist.
However, studies show that "small starts" predict eventual success better than initial enthusiasm or early progress. This phenomenon is known as the fresh start effect - taking the first step energizes us and bolsters motivation.
So focus on starting without putting pressure on perfection. Progress and course corrections will follow.
First, Do It: Embrace the MVP Mindset
Doing it = get the simplest MVP out.
A Minimum Viable Product (MVP) represents the simplest version of a product or idea that allows you to test, gather feedback, and iterate.
By embracing this mindset (just get something done - it's OK if rough, a prototype, a draft), you focus on progress over perfection, understanding that getting something out into the world is far more valuable than waiting for the perfect moment.
Expand Your Comfort Zone
Venturing outside one's comfort zone can elicit fears of failure. Leaning into discomfort not only builds confidence and skills, but research shows it makes us more receptive to learning. Recognize that fear is often the mind's way of urging us to grow. Don't let it stop you from progressing.
Then, Do It Right: Refine and Correct
Doing it right = fix correctness issues.
Once you've taken that first step and put your MVP out into the world, it's time to refine and correct. This stage is about learning from feedback, identifying areas of improvement, and making adjustments accordingly.
It's a chance to iterate on your idea, ensuring that it meets the needs of your audience or customers while aligning with your vision.
Cultivate Curiosity and Resilience
Meeting new challenges with curiosity and resilience makes venturing outside our comfort zone more sustainable and enjoyable. Cultivate curiosity about growth opportunities and your capacity to rise to them. Set mini-challenges to incrementally expand your horizons.
When facing inevitable setbacks, avoid self-criticism and tap into resilience - the ability to recover, learn and continue progressing.
Self-compassion, adaptability and maintaining perspective are key here. With consistent effort, you build confidence in your ability to start, stumble, learn and work toward mastery.
Finally, Do It Better: Strive for Continuous Improvement
"Doing it better = iterate towards an ideal end-state (e.g., make it fast)."
The journey doesn't end with merely doing it right.
The final step is to continuously improve, striving for excellence and growth.
By iterating towards an ideal end-state, you demonstrate a commitment to progress, ensuring that your product, idea, or project remains relevant, innovative, and successful.
Set New Goalposts
As you improve, have a clear idea of when you are “done” or update your goalposts. Elite athletes turn small gains into competitive edges via the aggregation of marginal gains. Identify areas of potential improvement and set measurable stretch goals, from increasing efficiency to enhancing user delight.
Overcoming the Greatest Barrier to Progress
"The greatest barrier to progress is not lack of resources or talent, but fear of failure."
Recognizing that fear of failure is the most significant obstacle in the pursuit of success allows you to confront it head-on.
By acknowledging this fear, you can focus on taking that first step, knowing that once the ball starts rolling, it becomes much easier to keep it in motion.
Remember that starting is more than half the battle. Don't wait until you feel ready, because the perfect moment may never come.
The Bottom Line
Rather than striving for perfect execution, embrace the power of starting - put forth an MVP, soft launch an initiative, or set a milestone. Progress begets motivation. By simply starting, you open the door to growth and innovation. The rest will follow.
Embrace the power of starting and then iterating until you're happy.
ALT First do it, then do it right, then do it better.
ZX🇨🇦 retweeted
The second question I asked @JeffDean:
Do you think it's necessary to have a long pedigree to actually start a startup in the AI scene?
At the time, I already had two first-author papers at top AI conferences with 3 years of undergrad left.
So I was seriously considering a PhD.
His answer was more nuanced than I expected:
> You don’t need a PhD - Practical, hands-on experience can teach you how to solve hard problems.
> Grad school is nice - it teaches you how to dive into a new area, figure out what the important problems are, understand what’s already been done, and connect ideas. Those are useful skills for almost any technical career.
> Building a company is different - You need to find something that’s important, matters to people, and is commercially valuable. Academic training isn’t necessarily designed to teach you that.
If I did want a PhD, he recommended doing it early.
Once you’re established in a career and earning a good salary, going back becomes much harder.
That answer made me rethink how I was thinking about a PhD at the time.
But the third question was the one I was most curious to hear Jeff’s answer to.
ZX🇨🇦 retweeted
The first question I asked @JeffDean:
How do you approach problems outside your area of expertise?
Jeff’s answer came down to three things:
> Find unfamiliar problems - Use the skills you're already good at in areas where you don't know much about the domain yet.
> Learn from people around you - Work with people who have different expertise. You learn from them, they learn from you, and together you can solve problems that neither of you could solve individually.
> Keep learning after school - Put yourself in situations where you're constantly learning on the job.
As Jeff was explaining this, I thought about a story I'd heard about him during COVID, from co-workers at Google.
Jeff was using Google Meet and thought it was painfully slow.
It frustrated him enough that he spent his weekend digging through the code.
The funny part: he didn't work on Google Meet. He didn't know the codebase. It wasn't even his team.
Then he found the bottleneck, changed a few dozens of lines of code, and made Google Meet significantly faster in just a few days.
That story made Jeff's advice click for me.
He didn't need to be the Google Meet expert. He just needed to be curious enough to ask, "Why is this so slow?" and keep digging until he found the answer.
It made me wonder how much technical pedigree actually matters, especially if you want to start an AI company.
That led me to my second question.
ZX🇨🇦 retweeted
The first autonomous agent cyberattack is an unprecedented event that deserves unprecedented transparency. Today we’re sharing everything we can: a full technical timeline, an interactive replay, and how we used an open model to defend ourselves, so defenders everywhere can learn from it and prepare for what’s next.
huggingface.co/blog/agent-in…
ZX🇨🇦 retweeted
Over the past few months, I've been dissecting how modern coding agents work.
Specifically:
Claude Code
OpenCode
Pi
Aider
And I documented everything I learned in a new open-source series…
It’s called “Building a Coding Agent From Scratch.”
Most people focus on the agent loop.
But that's only one piece of the system.
What makes coding agents work is everything around it:
Context engineering
Human-in-the-loop workflows
Durable execution
Permissions & sandboxes
Skills & subagents
AI evals & observability
The LLM is only one small component.
And the harness around it is what makes coding agents useful.
Over the next eight lessons, we'll build a production-grade coding agent from scratch in Python using:
@modal to host open-source LLMs such as Qwen3 and run remote sandboxes
Kitaru (by @zenml_io) for the durable runtime, so agents can pause, resume, and recover from failures
Opik (by @Cometml) to evaluate, observe, and compare agent runs in production
Pydantic AI for the bare-bones agent loop, showing just how little code makes up the "agent" (and showing how much engineering lives in the coding harness built around it)
And here's what you will learn inside the series:
1. System architecture
2. Agent loops & human approval
3. Durable execution & replay
4. Context engineering
5. Permissions & sandboxes
6. Skills & parallel subagents
7. AI evals & observability
8. Deployment
The first article will go live tomorrow in Decoding AI Magazine.
If you've ever wanted to understand how coding agents work under the hood, this series is for you.
Subscribe to get the first lesson in your inbox: decodingai.com/p/building-a-…
ZX🇨🇦 retweeted
“Loop engineering” is a hot buzzphrase after mentions of it by Boris Cherny (Claude Code’s creator) and Peter Steinberger (OpenClaw's creator) went viral on social media. Loops are now a key part of how we get AI agents to iterate at length to build software. In this letter, I’d like to share my 3 key loops, shown in the image below, for building 0-to-1 products. These loops guide not just how I build software, but also how I decide what software to build.
Agentic coding loop: Given a product specification and optionally a set of evals (that is, a dataset against which to measure performance), we can have an AI agent write code, test its work, and keep iterating until the code is bug-free and meets its specification. This idea of closing the loop took off around the end of last year, and it has been a game changer in enabling coding agents to work longer productively without human intervention. For example, over the weekend, I was building an app for my daughter to practice typing, and my coding agent could easily work for around an hour, using a web browser to check what it had built multiple times before getting back to me, without needing my intervention.
The engineering loop executes quickly. Every few minutes, the coding agent might build and test a new version of the software. I hear frequently from developers who are finding new ways to engineer more effective engineering loops. This is an active area of invention!
Developer feedback loop: In this loop, a developer examines the current product and steers the coding agent to improve it. Last year, a lot of developers (including me) were acting as the QA (quality assurance) function for our coding agents, manually finding bugs and then asking the agent to fix them. But with coding agents much more able to test their own code, the amount of time we need to spend on this function has decreased significantly. This allows us to make higher-level product decisions, such as what key features to offer, where the UI needs improvement, and so on.
The developer-feedback loop operates over time intervals between tens of minutes and hours — that's how frequently a developer might review a product and give feedback. In the case of the typing app, I changed my mind a few times about the visual design, what cat costumes she can unlock as she learns (she loves cats), and the user flow for a grown-up to log in and steer the child's learning experience.
When a developer has a clear vision for what to build, it is still a lot of work to translate that vision into a specification for a coding agent to implement. Further, after the developer has seen an implementation, they might update (or perhaps clarify) the spec to steer it toward what they want. If you find that the system repeatedly runs into certain problems, building a set of evals for the agent becomes useful.
AI-native teams are increasingly using AI to help shape product direction, for example, automating the gathering and analysis of usage data, summarizing written and verbal customer feedback, or carrying out competitive analysis. However, for pretty much all the products I’m involved in, I see humans as having a significant context advantage over current AI systems — we know a lot more than the AI system about the users and the context the product has to operate in — and thus humans play a critical role. Many people describe this human contribution as “taste,” but I prefer to think of it as humans having a context advantage, since that gives us a clearer path to helping AI systems get better. This also speaks to why this step can’t be automated: So long as the human knows something the AI does not, human-in-the-loop is needed to to inject that knowledge into the system.
External feedback loop: This includes a wide range of tactics like asking a few friends for feedback, launching to alpha testers, or putting the code into production with A/B testing. These tactics are usually slow, rarely taking less than hours and sometimes taking days or even weeks. This data informs the developer vision, which in turn continues to drive the detailed product spec, which in turn drives the coding agent.
With coding agents speeding up software development, more engineers are starting to play a partial product management role. For many engineers who are growing into this role, the hardest part is shaping the product vision and striking a balance between building (bridging the gap between vision and spec) and getting user feedback to evolve the vision. It is important to do both!
I will write more about how to do this in future posts, but for now, I find it encouraging that engineers are playing an expanded role (just as product managers and designers now do more engineering).
[Original text: The Batch]
ZX🇨🇦 retweeted
AI agents are booking travel, signing into websites, and acting on your behalf. That creates a new security problem: until now, letting an agent log in meant exposing your credentials to the model.
Today, with @AnthropicAI, we're changing that. 1Password for @claudeai lets Claude use your stored credentials to complete real-world tasks without your passwords or one-time codes ever reaching the model, its memory, or Anthropic's systems.
You stay in control and approve which credentials an agent can use. 1Password handles authentication behind the scenes.
Available now on Mac for business, family, and individual customers.
bit.ly/4bLP5EJ
ZX🇨🇦 retweeted
Something I have been thinking about: in the past, the best engineers I knew spent a lot of time automating their work in various ways. Better vim/emacs automations, writing lint rules to catch repeat code issues, building up a suite of e2e tests so they don't need to smoke test the app manually. These kinds of things were the highest leverage activities an engineer could do, because it multiplied their own output, which in turn meant they could build more things.
I think many of these automations have become even more important now. This is true for a number of reasons.
First, infra and DevX automation speeds you up. And if you are running an army of agents, each of those agents will be sped up also. More automation == more output per unit of time.
Second, moving things to code improves efficiency. Your agent could fix an issue every time it sees that issue happen, but that uses tokens and might miss cases. If Claude instead writes a lint rule, CI step, or routine, that class of issue can be fully automated forever. This is really what people are talking about when they talk about loops -- it's about automating entire types of busywork rather than solving them one off. This isn't a new idea at all. Engineers have been doing this for a long time!
Third and most importantly, automation makes it possible for others to contribute to the codebase more easily. Increasingly what I am seeing is engineers are contributing to codebases on day one because Claude can navigate the codebase for them, and that non-engineers are able to contribute to a codebase as effectively as engineers can. What gets in the way of both of these is domain knowledge that lives in peoples' heads rather than in automation -- the stuff you used to have to learn when ramping up. What has changed thanks to agents is the domain knowledge that can be encoded as infrastructure is no longer limited to what is expressible in lint rules and types and tests; it can now capture nearly all domain knowledge, encoded as code comments and skills and CLAUDE.md rules and memories. If I put up a PR for an iOS codebase I don't know and a code reviewer rejects it because it doesn't use the right framework, or if a designer builds a new feature and it gets rejected because it doesn't follow the right architectural patterns, these are failures of automation.
Every team should be writing the CLAUDE.md's, REVIEW.md's, skills, and docs that enable agents to productively work in their codebase with zero additional context from the prompter. This sounds crazy, and at the same time is a natural extension of the stuff engineers have always done: automate, and encode domain knowledge as infrastructure. As the model gets smarter and as the harness matures, this task becomes easier. In the meantime, it is on every team to look for ways to convert their domain knowledge to infra so that Claude can write code better, so that code review catches issues automatically, and so the next person working on your codebase can contribute more easily.
