Today, I'm launching a new AI course so you can copy the exact AI skills and workflows that I use to run my work and life. Since going independent, the most useful thing that I've done with AI isn't trying every new tool or vibe coding apps. It's building an AI system that proactively takes work off my plate and can work across any model or harness. Across 25+ bite-sized lessons, you'll: 1. Build an AI chief of staff to manage your inbox, calendar, and other tasks. 2. Set up an AI advisor to make better decisions about your life and career. 3. Automate workflows by chaining together reusable AI skills. 4. Make AI proactive with my best scheduled tasks for work and life. I've included 40+ copy/paste prompts for you to set all this up in an afternoon. In addition to the self-paced course, you'll also get: - $600+ in AI tool credits for my favorite tools - 16 ready-to-install AI skills - Monthly live workshops with me in a private Slack community of 400+ builders Get instant access to all of the above by signing up as a paid subscriber to my newsletter: Join by October 7 to lock in $150/year. After that, the price goes up to $200/year. Learn more at and thanks for your support!

my most important lessons on using voice input on agents talking with voice is auto regressive, like an LLM: - we speak one word at a time - we think and talk simultaneously - and we can’t easily take back what we said this created some interesting characteristics: - what we say first will influence what we say next. try starting your sentence with “the most important two things i want you to do are…” and you’ll find yourself cornered into a certain way of framing your message - we generally don’t have a ton of time to think while talking. and the faster we talk, the less thinking we do. this is why people who trained themselves to take pause are often more eloquent and perceived as thoughtful - we often perceive talking as higher stake than typing because correction is a lot harder these traits made voice input a terrible method for writing content. say you want to write a blog post, and using voice means you have to think through most things upfront, and randomly derail the content during talking due to choice of words, and can’t correct what’s written (unless you fallback to typing in the end) BUT.. when using agents, most of these traits don’t apply! agents can understand messy rambling just fine. they can understand your corrections if you said “no no scratch that, let me rephrase”. and they couldn’t care less if you take a pause so when using voice input on agents, you have to think of it differently from talking to real people or writing real content my advice: - don’t try to speak only the “final result”. speak your thinking process out loud this gives yourself a lot more room to reason through the concepts and arrive at something that actually represents what you want it also gives the agent more context on why you made certain choices you can even intentionally use phrases like “let’s think through this..”, or “here’s why..” to elicit more thinking - correct yourself often when you realized something you said earlier was wrong, you can either just stop the voice input and discard what you said so far, or start correcting yourself the good thing about agents is that they don’t judge. they are trained to never shame you for making mistakes or walking back your decisions once you started doing this more, you will perceive voice input as low stake, which makes it feel more natural and less demanding on the other hand, typing is text diffusion - you think in your head and type out the results, and you constantly refine what you already wrote it’s great for writing polished content, or precise text such as code. but it’s not as good for prompting agents because all your thinking tokens were hidden so the agent only sees the final “what” but rarely the “why”

youtube discovered i decided to start eating healthy, so it decided to recommend a michelin chef cooking ramen to me at 11pm... watching the first half i was like "i do that too" the second half just showed me why i'm not a michelin chef

how to set the right reasoning effort level i get asked this a lot, and i learned that because the underlying mechanism is not obvious, a lot of people don’t have the right mental model to think about how to use the knob so a quick crash course here: reasoning effort level in mainstream LLMs today means a “ceiling”, not a “floor”! setting a high effort level does NOT mean every prompt you send will use a lot of thinking tokens. you can try this very easily - launch your agent in an empty directory, use high reasoning, and say “hi” - unless your harness is totally messed up, the agent will respond almost instantly how does that work? the most typical implementation is that in post-training, the models receive a penalty for how many thinking tokens they use. how strong the penalty is depends on the thinking effort level that’s set at low reasoning, the penalty is very strong, so any long thinking traces get punished into oblivion at high reasoning, the penalty is weak so longer thinking traces don’t get punished too much, and are allowed to survive more often as long as they get good results the most important thing to learn there is - thinking tokens is always a penalty, not a reward. there’s never a rule that says “at high reasoning i’ll punish you if you think too little” - this is because more thinking tokens always means higher cost which is always a bad thing and should not be rewarded so with that in mind, you should think of reasoning effort as “how much are you allowed to think”, not “how much do i want you to think” setting it at low means “for this task you are not allowed to think much”. you would want this when 1) the use case requires low latency; or 2) you already know the task does not require additional reasoning setting it at high means “you are allowed to think more when needed”. the model will still try to think as little as possible according to what its training data teaches it. you want this when 1) the use case allows slower response; and 2) you are not sure how much thinking is needed for this task for some models, “max” is special because it forces thinking mode to be on, which forbids “zero thinking”. i don’t have a full inventory for which models do this vs not. and i generally avoid this mode because i don’t think it’s a good idea to say “every request has to have thinking tokens no matter what” practically speaking, i mostly only use two modes: - medium, when i already know the task is well defined for example, implementing a spec planned by a highly intelligent model. i avoid low because 1) i rarely require low latency; 2) even when implementing well defined tasks sometimes the intermediate context is still ambiguous this is also my default for firstmate because highly ambiguous problems are typically handled by crewmates, and i do want firstmate to be fast - xhigh, whenever the task is not well defined yet. this is most often used for planning and investigative work. i do this to give enough room for the model to decide how much thinking it needs hope this is helpful!

alright! as the dust settles around this insane week of model releases, i've stabilized around a new model line-up so sharing here for reference this time my approach is a bit more structured. i've bucketed various LLMs into a few categories: 1. interactive orchestrators these are models that i use as firstmate (and second mates), because they are fast, efficient, pleasant to talk to, intelligent enough to understand my intent, and have good enough judgment to steer the crew around it for me, this bucket is opus 5.5 and grok 4.7 viable budget alternatives when my subscription quota runs out: muse spark 1.3, deepseek v4 flash 2. premium intelligence these are models that i only use for highly ambiguous or creative tasks that i decide to truly need the extra intelligence and justifies the cost i also use this bucket to handle escalations - when crewmates started arguing with each other, when a review-loop started spiraling out of control, when a simple change somehow ended up with a giant PR - i call these models to untangle the mess for me, this is currently gpt 6 astra and fable 5.1 (or opus 5.5 when fable quota is tight) 3. planners these models are the ones i trust as default for planning new features and investigating complex bugs. they would produce a spec that get implemented by a cheaper model i almost exclusively use opus 5.5 for this right now because of its incredible ROI, whenever i don't need the premium intelligence 4. implementers these are efficient workhorses that when given a well defined spec they can produce solid implementation i use opus 5.5, gpt 6 sol, and grok 4.7 for this right now, and again the budget options: muse spark 1.3 and deepseek v4 flash (i'm sure there are many other viable alternatives as well - i just haven't got enough time to try them out) among these models i currently find sol to be the best adversarial code reviewer (i haven't tried astra for this, as it's a bit too expensive to run at such high volume) 5. trivial fixers these are the models i use for extremely trivial changes like a one-liner fix or config change. they really don't need much intelligence because often times what needs to be changed was already defined for these i use gpt 6 luna and again the budget options if quota is tight the way i actually make use of the categorization is that i told these routing preferences to firstmate, and it can then help me route the right task to the right model (which is now made very efficient because of Jev) one last thing i'll point out is you can see how versatile opus 5.5 is in my line up - it can do pretty much everything! this is the first model that spanned across almost every bucket in my setup, which is quite a massive advantage because it means if i have to choose only one subscription it would have to be anthropic at the moment

Hey captain @kunchenguid, look what I built for Firstmate on @fabienpenso’s Herdr-gpui edition 👀 Captain’s Deck: live Kanban - Charted Next → Underway → Captain’s Call → Awaiting Merge → Landed. Answer Captain’s Calls from the board.

fable 5 was good. opus 5 and sonnet 5 fumbled astra 6 was good. sol 6 and luna 6 kind of fumbled now opus 5.5 is good. i'm guessing sol and luna 6.5 will improve too it seems openai is a few months behind?

the definition of "slop" is not constant something we may see as impressive today will be seen as slop a few months down the road i think the constant part is - if something can be one-shot by AI, it is slop because if you can one-shot it, millions of others can too. and when that happens, people associate it with being low-effort and worthless this is one of the reasons that fully automated software factories are not going to work because by definition what they produce is always slop the ones that do work are those that put humans at the center and are designed to help amplify human craftsmanship rather than trying to replace it

Using @kunchenguid firstmate with Opus 5.5 is an incredible experience. Especially helpful for those with significant responsibility during the day unable to manage agents, firstmate keeps the work going without you having to babysit.

just took gpt 6 luna for a spin, and… it’s a very weird model 1. it’s insanely cheap, even with the point below considered 2. it’s extremely slow, not in terms of time to first token or toks/sec, but how many turns it takes to get something done 3. it eventually does get shit done.. this created an interesting condition - using it interactively feels totally unusable because you have to wait for many turns before a useful outcome comes back i tried it with high reasoning as firstmate and it straight doesn’t work, because by the time it finishes the current turn there are already two more turns worth of events piled up for it to process. it cannot keep up and literally will never finish but if used as a background workhorse, and you don’t care that much about e2e latency, then it’s _extremely_ cost-efficient and capable - nothing else even comes close to this ROI when the labs brought us “fast mode” which makes the models faster but costs more, i jokingly said i actually wanted a “slow mode” turns out luna is the slow mode

since it’s been a good day for anthropic with a strong opus 5.5 release, i’m going to highlight one more thing that may not be obvious across xai, openai and anthropic - 1. anthropic is the only provider that does not charge 2x for long context requests even at 1M context, anthropic charge at the same flat rate, while openai charges 2x above 272k tokens, and grok charges 2x above 200k 2. anthropic’s latest models have ridiculously low pricing for cached read. see chart below most coding sessions have 95%+ cache hit rate, so this difference is massive my hunch is that eventually this will even out, but for now, this is a very material difference that’s easy to overlook

ok i've used opus 5.5 enough now to have an informed opinion and i just want to say - hallelujah!!! opus 5.5 solved SO MANY problems. let me list them below in increasing importance 1. verbosity is now used more appropriately i still see long responses, but only when it's justified. there's much less fluffy "claude speak". it's now mostly real talk 2. the behavioral problems in opus 5 are fixed i haven't got a single "honest mistake" yet. unlike opus 5 which likes to jump into conclusion, make a wrong move, and then admit its problem and take more turns to fix it, 5.5 just does the right thing it also does it FAST - this made a visible difference when using as firstmate, and was one of the reasons i really liked grok. now i would say opus 5.5 is on par here 3. MOST IMPORTANT - cost i can now finally ignore the 50% cap on fable, and fully utilize the subscription value with just opus for pretty much everything i still find myself wanting fable for some ambiguous, creative prototyping work where i kind of want to say "surprise me". i tried opus for one of those things earlier and it didn't surprise me.. but for literally everything else, 5.5 is just nailing them without a miss so far. so for the first time in a long time, i finally stopped using opus 4.8 as my firstmate & all my second mates, and switched that to opus 5.5 overall, very impressed - now i know anthropic did know opus 5 was bad despite not willing to acknowledge it. :) they knew exactly what to fix, and they got the job done

ok i think xai, anthropic and openai have been in the same space for so long that their cycles started to synchronize it's quite something to see grok 4.7, opus 5.5, gpt 6 sol, terra, luna all coming out at the same time

day 1 observations for grok 4.7 ignore the reports that say “it’s terrible” and the only thing they reference is a public benchmark. the same benchmarks told us opus 5 was better that fable - they are useless also ignore the reports that compare models with 3d games - that’s not real work. it's made for attention on social media i used grok 4.7 for a whole day as my firstmate, and it has been a really solid model with visible improvements over 4.5 (i'm ignoring 4.6 because 4.5 has been working better in my experience) key differences with 4.7 - 1. it follows system prompt very, very closely i noticed firstmate showing many new behaviors that i've never seen before, such as asking me to name specific red CI checks that i'm ok with bypassing, and refuse a simple "yolo" instruction i traced it and it's indeed how i instructed it in firstmate's system prompt, but none of the other models followed it closely enough to make this behavior visible - grok 4.7 is the first to pick that up there were a few other similar examples as well. so to me this is a clear behavioral difference 2. it's very "stable" if you've used astra then you know what a "spiky" model is. it can have some genius moments but you occasionally also wonder "how could it be so dumb and doesn't get me". grok 4.7 is the opposite of that throughout the whole day so far, i'll be honest i haven't get a "wow this is absolutely genius" moment yet, but grok 4.7 has been very steady with no big surprises. its behavior feels predictable, which does help it gain trust from me quickly 3. it's a conservative model it doesn't like to take actions without asking, and would explicitly say so this is a bit of a double edged sword, because it means i sometimes have to state the obvious "yes i do want that", but in hindsight a lot of those cases are indeed a bit ambiguous and i may not have preferred the model to just move forward without my confirmation 4. it's a bit slower and costs more than 4.5, visibly turns are taking a bit longer and my quota is draining at a visibly faster pace. i haven't quantified exactly where this is coming from yet so overall, i think it's showing some clearly different traits, and i mostly like the changes. i'm going to keep it as my primary firstmate and observe more if you've been using it, what qualitative insights have you gathered from real usage so far?

I bookmark a lot repos so i can come back and review them later. Most of them i never end up using. First mate is absolutely fantastic. Five stars. Would recommend. Life changing. thx @kunchenguid

IIRC grok 4.7 is a bigger model than 4.5 and 4.6, but it kept the same price - this likely makes it the best day-to-day LLM there is trying this out for the rest of the day and will report my experience!

when your agent tells you "i'm not a lawyer. you need a proper legal counsel to sign off on this", how many of you just spin up another agent to do the legal review, and pass it back to the first agent saying it's from your lawyers? asking for a friend