seer

It’s not equal amount of time though.

You don’t need to support all the other customers, and you also have the exact apis and their usage.

Telling Claude to “reimplement” something is vastly more achievable than trying to spec and research and develop a totally novel idea.

And it gets much worse for the “open core” projects - you can literally get the core and vibe code all the “premium” features and don’t pay the original creators a dime … while still requiring a lot of skill to do, it is vastly more tractable than it used to be, shifting the “is it worth the time and money” point quite a lot.

My prediction is we will see a lot less open source or open core companies in the future, people are now guarding their IP much more.

Ahh, it was nice while it lasted. I’m already thinking how I will be telling my grandkids about the good old time of GitHub and open source and oss social interactions, while stroking my white beard…

Yeah but depends how you use it - with superpowers and it’s prevalence of splitting things into smaller focused subagents - this could seriously reduce costs …

I wish my company gave me more options than just using Claude to test these things out

Getting on and off fable this week has been quite interesting. For my personal work stream (big terraform monorepo, hundreds of states) I’ve using mostly superpowers to do heavy / quality work. But with fable, I tried just telling it what to do, and it produced roughly the same results without a big structured back and forth that I was accustomed to.

Then after using up all my fable allowance I figured let’s see if opus can actually work without superpowers, and no, it was all over the place doing weird things.

Thing is, superpowers produces meticulous specs and plans as a byproduct of its work, which is very useful for switching between work trees, stoping / resuming work by different people.

But to do that in Fable you have to spend way more tokens than it’s reasonable. You get similar quality result, but without the specs in between.

I’m not super sad that I’ll have to go back to opus though, with superpowers it was Fable but more structured. But I will miss the banter though - Fable is amazing for brainstorming big underspecced features.

I’ve found that Claude is really good at picking up tone of voice in prompts / queries.

If I go “find issues in this code” it will hallucinate some, but if I say “can you check the recent change, there might be some things that introduced regressions, maybe?” Then it will be more cautious.

Also especially fable but opus too can talk back and advise you against going into a direction it thinks unwise.

And I’ve had much more success in clearing out why I think that is a better approach or asking it to clarify itself, as if if I tell it my assumption, sometimes it self corrects and starts doing what I needed in the first place, it was just coming at it from a different direction before. For example assuming I don’t care about cost and providing “the best solution” or trying to make something reusable where what I needed something quick (or vice versa)

It really is best to think of it as a gradient plane where it might get stuck in local minima, or you can prime it to “teeter on the edge” and able to flow into different directions.

Ha! It’s silly easy for me to bum in the “premium” tier - the highest one my employer is able to select for me.

With proper skills and validations, it’s quite easy for me to spin out a Claude instance and keep it running in the background for every idea / problem / bug etc.

Like for example when a new request comes in, or when I have an idea for refactor / improvement, I brainstorm, weed out all the uncertainty and details, then just create a plan and let if follow it (using sonnet / haiku for execution)

I would have 4-5 simultaneous instances running - and all of them produce valuable results.

But then 2-3 days into a week I hit my weekly limit.

And the company can’t pay more for me within Anthropic’s enterprise structure, except for the “extra” api costs. Which itself sounds quite silly to me.

I’ve resorted to running the brainstorming and planning with Claude, but have other tools / companies execute the actual implementation… doesn’t work as well but what can you do…

Yeah but I think going back to hand writing bespoke code is not coming back, the genie is out of the bottle.

But we could build much better tooling around keeping the agents honest. The problems you are describing are absolutely real and I see them every they.

One friend of mine had almost a mental breakdown when he just went ahead and drilled a bug producing Claude to the point that it itself admitted it was “a piece of shit”. He knew that arguing with an LLM agent is more than useless, but it was cathartic for sure.

When I encounter a situation like this I always go down to - have I done everything I could to catch these errors in my automated validation, and update it as needed.

Agents are also more than happy to spend tokens refactoring, once you have such a test harness be good enough, producing successively better and more general abstractions is quite easy.

The old rule of thumb of “make it work, make it fast, make it pretty” still applies , just with much much faster iteration speed.

It seems with agents people have forgotten the last 2 steps since they produce a _working_ solution, and it might be hard to justify spending time “cleaning it up”, but this still remains essential.

I just eat double portion at dinner, and then nibble on snacks before bed - I haven’t had breakfast in decades, then since I moved to India, with the carby nature of the food it was hard to stay in shape with 2 meals, so I decided to try and skip the lunch too. With fun work it is actually quite easy, and babysitting 4 claudes and helping out colleagues is very entertaining.

Now I either do gym before dinner (heavy exercise) or social dance after.

I’ve been given a lot of advice how I “should” be structuring it - like “don’t eat too much before bed” or “never eat before exercise” … but I haven’t had any issues with what I’m doing so far (~2 years)

Exactly - I had switched to a one meal per day setup and have been mostly following it for a few years.

Then after a routine “heart health” check all my indicators were super out of whack - the doctors thought I was on my deathbed - but I am perfectly happy pain free, in shape, physically active person…

Then _i myself_ had to dig into all these tests and figure out that they were measuring the wrong thing - since they try to time where your body is “just about to eat after a fast” - normally for most people in the morning before breakfast, but since my first meal of the day is usually around 20:00 - my body had adopted to have higher levels of various things just to stay on top of my lifestyle choices.

Anyway I had to educate some doctors since they haven’t really had a case like mine, so they weren’t thinking critically of how to interpret the results…

I imagine an automated test _could_ take these things into account with large enough dataset, but it would need to do a lot more reasoning than statistical correlation.

I do believe current sota models should be good enough to come to the correct conclusions with the right harness though.

Sad they didn’t finish it. The later books seemed like they were written “to be movies/series” with less political stuff and more visuals / speculative sci-fi. Plus the story has an actual half decent conclusion.

To me the plight of Hollywood is “forever shows” where the writers start without knowing where it would end, so shows never really “end” they just slowly fizzle out without ever getting to the end.

I simply loved “the good place” for having such a powerful conclusion.

I really liked reading through the Mars trilogy. It imagined a world where AI is used for fluent effortless translation - local languages get a renaissance since now you _dont_ need a lengua Franca, everyone just speaks what they like the most, and can understand everyone else. Much more “flavour” to human interaction.

Also ai makes things just resource constrained, not labour - whatever you imagined, you could make happen, just needed to “talk to an ai” about it. Lots of terraforming Mars / Venus in that book were imagined like that.

But it also analysed the social / political / behavioural aspects of it. Places that had to preserve old power structures - aka US/Europe/China - got engulfed with mega corps controlling everything etc.

But Mars - where people had enough freedom to imagine something different, came up with political/financial structures to incorporate all of that, and thrived.

I think it tried to play the card of “if US was being created right now - what would its ideals be” If you had a huge tract of land that was “free” and nobody (powerful enough) claiming it, and a population that didn’t yet have strong allegiances and could be persuaded to band together, what would AI, tech and all these years of progress allow us as humans to achieve politically.

Which also makes you feel kinda sad for the US in that world - it is the old rusted power center that can’t innovate and is stuck in the past…

Now it’s only sci fi of course, but it was quite interesting to imagine a world where AI gets smarter and smarter but never reaches that “sentient” threshold. I think the whole trilogy aged incredibly well all things considered.

This article kind of misses the point - of course everyone’s average day is … average. But locals don’t spend all of their days like this, sometimes (once a week/month/year) they would do something fun, or they want to.

You asking them for advice or for them to show you around might push them to do something fun themselves, which they haven’t done in a while. But they have a lot more local context about what _might_ be good to explore or not.

They also know people - they themselves might have average days, but everyone knows that fun person that is the social glue that does all the fun stuff they can direct you - 7 degrees of separation and all that.

And lastly sure - treat the locals ideas with a grain of salt - I never do _exactly_ what the locals tell me, but it is another data point to make your own plans.

When I travel I like to make huge holes in my plans - uncharted time for me to fill in when I’m at location - from local sources or just doing the research then and there. It has always been more natural and interesting to do the sight seeing planing at location, so you can adjust and correct anyway. I guess have adopted the startup mentality of start small and iterate even for my travel experiences :)

But isn’t AI doing the same thing to project management as to coding?

PMs can now cross reference and organize tickets with just a few keystrokes. Organisational knowledge, business knowledge, design systems and patterns, etc all of it is encoded in LLM consumable artefacts. For PMs it is the same switch - instead of having to do it by hand you direct lower level employees to handle the details and inconsistencies and you just do vibe and vision.

When all of the pieces successfully connect and execute reliably, what is left for humans to do? Just direct and consume?

And AI companies with their huge swaths of data are soon gonna be in the situation of being able to do the directing themselves

Thanks for the recommendations - I’m also big fan of 3blue1brown and PBS science, but as a recent dad am on lookout for content for my son to watch when he comes of that age - he’s just 1month now, hopefully by that time AI has not enshittyfied everything

I’ve noticed that agents almost always fail at the planing vs execution stage.

I follow the plan -> red/green/refactor approach and it is surprisingly good, and the plans it produces all look super well reasoned and grounded, because the agent will slurp all the docs and forums with discussions and the like.

Trouble is once it starts working there would inevitably be a point where the docs and the implementation actually differ - either some combination of tools that have not been used in that way, some outdated docs, or just plain old bugs.

But if the goals of the project/feature are stated clearly enough it is quite capable of iterating itself out of an architectural dead end, that is if it can run and test itself locally.

It goes as deep as inspecting the code of dependencies and libraries and suggesting upstream fixes etc. all things that I would personally do in a deep debugging session.

And I’m supper happy with that approach as I’m more directing and supervising rather than doing the drudgery of it.

Trouble is a lot of my team mates _dont_ actually go this deep when addressing architectural problems, their usual mode of operandi is “escalate to the architect”.

This will not end up good for them in the long run I feel, but not sure what they can do themselves - the window of being able to run and understand everything seems to be rapidly closing.

Maybe that’s not super bad - I don’t exactly what the compiler is doing to translate things to machine code, and I definitely don’t get how the assembly itself is executed to produce the results I want at scale - that is level of magic and wizardry I can only admire (look ahead branching strategies and caching on modern cpus is super impressive - like how is all of this even producing correct responses reliable at such a a scale …)

Anyway - maybe all of this is ok - we will build new tools and frameworks to deal with all of this, human ingenuity and desire for improvement, measured in likes, references or money will still be there.

Yes, but the YouTube ed channels are such a treasure in and of itself. We had the “tech” to produce content like this for almost a century, but it took the Internet and democratization of content creation to come up with gems like smarter every day, veritasium, extra history, etc

My fear is that this is also being reshaped with ai, mostly for good now but I feel like the personal touch and passion of these creators is being diluted with the advent of generated content.

Maybe we are in a valley of the uncanny valley and the ai tools will become so good that they can successfully translate someone’s passionate vision faithfully, then it could be another renaissance.

What the author is missing is that in his decision to limit the use of LLMs in his work, he omits the part where he “can”. E.g. he is resourceful and accomplished enough to be able to do the work he desires with no LLMs - but most people actually can’t. There are whole swaths of people software engineers that don’t write tests because “it slows them down” but they have never learned how to write testable code. And when thrust into an environment where they need to learn quickly - they don’t really have a way not to use ai, if they don’t someone else will, and take all the credit.

Learning how software is built is hard and gruelling work, and you need to constantly invest in yourself. Trouble is there is no time left to “go back to basics and learn FP” for example, because you also need to keep up with all the new LLM stuff happening on top of that.

It is easy for us who already have the foundational knowledge to be able to step back, take the wheel and try to do it ourselves, but plenty of people simply don’t have that option.

And I expect this trend to deepen and broaden. There will definitely be a lot more “witches” than actual engineers.

I think the right term for highways or most other car roads is “car sewer” - you need very specialised equipment to navigate them, they are deadly, smelly, loud and unpleasant. One of the worst environments humanity has produced.

Yes they ship people around somewhat fast. Slower than possible with other methods, and the cost is incredible - economic (much more expensive per passenger than almost any alternative), political (they inherently divide people, dehumanise and make people never really share a public space), health - they reduce lifespan by both lowering living quality as well as directly killing a staggering amount of humans per year).

And we have learned how to build better places for humans that do not need these coffins on wheels - if you visit any European capital, and most Asian ones - you will see environments built for humans, not cars - soo much nicer.

So cars as a technology have definitely not been beneficial to humanity overall, but it has been somewhat useful to some.

I think strongtowns were very good advocates of what places in America could like if you look beyond cars. I personally like the “not just bikes” channel though.

I had fun “hacking” my router that turned out to be just unzipping the file with slight binary modifications, it was so simple in fact I just implemented it in a few lines of js, even works in the browser :-D

https://ivank.github.io/ddecryptor/

Capital in the 21 century, how to win and influence people, Sidhartha, meditations by Marcus Aurelius, the mars trilogy, the nurture revolution.

These are off the top of my head.

The Catholic Church thing - yea that was quite unexpected for me, and apparently accidental for the church too - the basic premise was - they banned cousin marriage, and heavily enforced it throughout all of society - kings to peasants - this drove people to move around and settle outside of their home towns, driving up individualism and just changing the way our brains work on a neurological level - we have always been a close nit kin social structure animals.

The e book explains it quite well with tons of historical data, neuroscience, comparisons with different countries, continents and social structures.

It got me to “understand” India on a much deeper level since I moved here from Europe, and not get pissed off at people for “not thinking things through”. But also appreciate how small and consistent things can drive profound changes. Also how did china/ussr speed run the Industrial Revolution so quickly - spoiler alert - they copied the same “ban cousin marriage” thing

“The weirdest people in the world” - has a very good roots cause analysis of all this.

Basically banding into groups and guarding against outsiders is the default human behaviour. It just works that way if you do a game theory analysis of our social structures. They usually don’t scale too well, but that’s what we revolved to do as social creatures.

It’s actually and very counter intuitively the Catholic Church that lead us to individualism, common laws, nationalism, even the Industrial Revolution and the scientific method.

It sounds bizarre but if you follow the historical logic, in a round about way it has paved the way for the modern world, which the rest of human civilisation was forced to adopt, either to compete or at gunpoint.

There are few books I read in a year that change the way I look at the world, “The Weirdest people in the world” was definitely one of them.

We did a similar Claude code mandate a few weeks ago.

Motivation was people being so allergic to tests and automation, that making them use superpowers produced better code, but also started adding a test pyramid.

The mandate was actually phrased in a way that you must produce industry standard code, and if you struggle with it you can use cc to bridge the gap.

Honestly I worry that this way devs will produce higher quality code, but will not understand why, how to measure the “quality” and steer towards it themselves.

At this point though the founders were pretty adamant with the code quality and lack of tests so this seemed like a reasonable way for the company, and I am curious to see how such a mandate affects code, deliverables and individual’s knowledge.

So far it seems to be working as intended, but it is early days.

IRGS domestic propaganda has always been that US is a military murderous malevolent regime, mercilessly going after their land and their children.

With just a little bit of propaganda spin, or even without it, US just proved to the entire Iranian population that IRGS was right all along.

This should strengthen or even harden their regime as they will have new generation of hardliners join the movement.

This is like 1930s Germany kinda thing. Who won or lost is semantics at this point, the regime is free to spin it any way they want, and will have quite the support to do it.

Is this the “city experience” in general or specifically for the United States? It famously has very poor urbanism so might not mean the same as in Europe for example.

I have grown up in rural Russia in the 80s and that was also similar - a forest started 50m from our house and I would just get lost there from time to time - not fun for my parents but magical for me.

Then we moved to the middle of a European capital city (Sofia) and I _still_ had almost a forest right next to the apartment block we used to live in - enough of a forest that as a 10yo kid I could find a nook to build myself a small hut with a burning fireplace inside it and nobody complained.

There are plenty of big European cities that are 10-20mins short unsupervised trip to a wilderness that a kid can do.

For example - Valencia has an uninterrupted bicycle highway that gets you from the city center to a wilderness preserve and a beach in less than an hour cycling.

To me all of these nature vs city laments are just US car dependency. Cities don’t have to be this way at all.

Is it worth the price of energy sovereignty though? You are not just buying electricity, you are buying future independence. It might be worth it if you factor that in.

And don’t forget all the other expertise that comes from being a country that is able to build reliable nuclear reactors. China is _the_ production superpower not just because it can build x or y, it’s because it has all the supply chains to be able to do it at scale.

If a country invests into that expertise - you get a lot of very capable engineers, a lot of tech and supply chains to deal with making it all happen, again and again, at scale. That in itself would be something that can offset the raw “price” of a single reactor, though it is very hard to quantify.

Like how much has USA actually lost by relinquishing its historical role of guarding international trade? Maybe it won some independence, but maybe the upstream effects to its economy would be bad?

We don’t know for sure about nuclear, but when a similar scientific project was put on a national scale - the space race - USA got silicon valley out of it.

Also it has interceptors, which allow you to build easily reusable pieces of code - loggers, oauth, retriers, execution time trackers etc.

These are so much better than the interface fetch offers you, unfortunately.

Haven’t seen the movie yet, but the book is definitely one of my all time favourites, so I would recommend reading it regardless of the movie.

The way the book is structured there is only one big reveal that would be spoiled by the movie, but I don’t think that was the most interesting thing in the book anyways, it was all about engineering, the scientific method and all that, and I think that will still hold before or after watching.

The one big exception I’ve found to “read the book first” advice to me has been “the expanse” there the books and the series were so great that they sort of complemented each other, and the advice there is “definitely do both”. I was reading the books and watching the series in parallel - side by side.

I do hope Hail Marry is like that…

MAD

If they strike desalination plants, Israel/us can do the same … really mass casualty event could follow.

And they might, at some point the Iranian gov might feel desperate enough to be like “fuck it, we have nothing to lose” … Dubai could end up with a lot more graves.

Almost all of their water comes from these plants, and humans can’t survive without water for more than 3 days …

There are reserves/stores sure, but how long will they last, and which part of the population do they cover? In a week you could have thousands of civilians dead on both sides.

So MAD keeps things in check.

I think this is whaly Iran has invested so much into rockets - they are very ineffective at providing decisive military victory by themselves, but without them, Iran will be at Israel’s mercy, and they have proven to not possess that in great amounts lately

They address this in the docs - it is meant to make authoring the content easier for LLMs since that is easy for them to write.

It still uses MJML for the actual templates, but it is a translation layer between markdown and the template itself.

If you need to author a lot of emails with LLM this does seem like it would be a great fit.

Isn’t a “local emulator of cloud services” kind of the perfect project to be vibe coded? Extremely well documented surface to implement, very easy to test automatically and prove it matches the spec, and if you make some things sub optimal performance wise, that is totally fine because by project will not be used in a tight loop anyway - e.g. it will just need to be faster than over the network hop plus the time it takes for the cloud to actually persist things. This can just need to do this in ram and doesn’t need to scale.

So I’m shocked cloud providers haven’t just done this themselves, given how feasible it is with the right harness

I think what they meant is you “can save 10 hours of planning with one hour of doing”

And I think this has become even more so with the age of ai, because there is even more unknown unknowns, which is harder to discover while planning, but easy wile “doing” and that “doing” itself is so much more streamlined.

In my experience no amount of planning will de-risk software engineering effort, what works is making sure coming back and refactoring, switching tech is less expensive, which allows you to rapidly change the approach when you inevitably discover some roadblock.

You can read all the docs during planning phases, but you will stumble with some undocumented behaviour / bug / limitation every single time and then you are back to the drawing board. The faster you can turn that around the faster you can adjust and go forward.

I really like the famous quote from Churchill- “Plans are useless, planning is essential”