I would be interested to see how people would apply this working as a coding assistant. For me, its application in solutioning seem very strong, particularly vibe coding, and potentially agentic coding. One of my main gripes with LLM-assisted coding is that for me to get the output which catches all scenarios I envision takes multiple attempts in refining my prompt requiring regeneration of the output. Iterations are slow and often painful.
With the speed this can generate its solutions, you could have it loop through attempting the solution, feeding itself the output (including any errors found), and going again until it builds the "correct" solution.
I basically did this with aider and Gemini 2.5 few days ago and was blown away. Basically talked about the project structure, let it write the final plan to file CONVENTIONS.md that gets automatically attached to the context, then kept asking "What should we do next" until tests were ready, and then I just ran a loop where it modifies the code and I press Return to run the tests and add the output to prompt and let it go again.
About 10 000 lines of code, and I only intervened a few times, to revert few commits and once to cut a big file to smaller ones so we could tackle the problems one by one.
I did not expect LLMs to be able to do this so soon. But I just commented to say about aider - the iteration loop really was mostly me pressing return. Especially in the navigator mode PR, as it automatically looked up the correct files to attach to the context
It has been interesting what both groups of 'yes' and 'no' chime in here. Personally I am on the side of 'no' but for a rather simple reason. I ask myself the following question:
Why spend time being good at something you don't care about being good at any more?
It is purely a personality thing however for me I would like to continue moving up the career ladder and you rarely see CTOs, VpEng rolling up their sleeves and sifting through CloudWatch logs. I want my focus to be on working the skills associated with those roles.
As a people manager that works with many incredibly capable engineers that are aspiring to be managers, I share with them this advice, 'excellent engineers compound their value by making other engineers excellent. It's far more difficult to do that when you are writing code.'
I have posited a similar idea with some of the people I work with. The issue of having complex, multi-step tasks be completed successfully has already been solved. You don't heavily invest in having one single expert for your business to solve all your problems. You build a team. Multiple specialized experts working in unison to achieve a shared outcome. Some people work on the task simultaneously, others sequentially. All with a specific purpose associated with the goal.
These assets are horizontally and vertically scalable based off skills, quality, or performance required. An efficiently designed AI architecture I believe could do the same. Its not mixture-of-experts as you aren't necessarily asking each model simultaneously but designing and/or having the system intelligently decide when it has completed its task and where the output should travel next.
Think of a platform where you had 'visual design' models, 'coding' models, 'requirements' models, 'testing' models, all wired together. The coding models you incorporate are trained specifically for the languages you use, testing the same. All interchangeable / modularized as your business evolves.
You feed in your required outcome at the front of your 'team' and it funnels through each 'member' before being spit out the other end.
I have yet to see anyone openly discussing this architecture pattern so if anyone could point me in that direction I would thoroughly appreciate it.
As a reasonably experienced programmer that has watched Andrej's videos the one thing I would recommend is that they not be used as a starting point to learn neural networks but as a reinforcement or enhancement method once you know the fundamentals.
I was ignorant enough to try and jump straight in to his videos and despite him recommending I watch his preceeding videos I incorrectly assumed I could figure it out as I went. There is verbiage in there that you simply must know to get the most out of it. After giving up, going away and filling in the gaps though some other learnings, I went back and his videos become (understandably) massively more valueable for me.
I would strongly recommend anyone else wanting to learn neural networks that they learn from my mistake.
For me 3brown1blue series: https://m.youtube.com/watch?v=aircAruvnKk was an excellent introduction that made Andrej's videos understandable. Then I did 3 first chapters of fastai book, but found it too high level, while I was interested in how things works under the hood.
Going through Andrej's makemore tutorials required quite a lot of time but it's definitely worth it. I used free tier of Google Colab until the last one.
Pausing the video a lot after he explains what he plans to do and trying to do it by myself was a very rewarding way to learn, with a lot of "aha" moments.
I think it's fair to say when you hit the hundreds of millions of dollars mark the diminishing returns for making things happen faster have well and truly kicked in.
Perhaps the only benefit would be extra computational power yet I would struggle to understand the benefit of jumping from 500 million to 5 billion with such short timeframes.
The ability to truly not think about training run costs, throw random things on the wall to see what sticks. 10x resources is definitely a competitive advantage in LLM training.
People will under-deliver when there is no clear outcome or agenda for their work. Within the office the pressure of contributing to work is provided because others can see what you have on your screen and if its not "work". You contribute because you have to, not because you feel that your outputs are moving the needle in a positive direction.
When you WFH that changes ('aint nobody watching your screen but you) but the underlying problem still remains. The team are not working towards a clear goal that they understand and want to achieve. When you provide that, the team will always contribute effectively because its interesting and importantly allows them to feel like their work means something.
WFH productivity is not the problem. Managers providing worthy work is.
I'd argue that it's almost never correct to drive the speed limit. In the US at least they are almost universally slower than the natural speed of the road by 5-10 mph and everyone speeds a bit all the time.
This is a bit preposterous. First off, do you know how speeds are set? Civil engineers build a road, and then figure out the 85th percentile speed for that road based on sample data collected. [0]
That's a couple things:
1. Not based on any actual design criteria for the road
2. Assumes that 85% of drivers are somehow more knowledgeable than the engineers who are supposed to be designing our roads and transportation.
The claim that you should be always able to go 5-10mph above the limit seems to be a mythos invented in and ingrained in the fabric of the American psyche. Better approaches exist than to just pick a popularity consensus from a sample of random drivers in a study (who are predisposed to going faster, often because the study is done before full levels of traffic set in).
> The claim that you should be always able to go 5-10mph above the limit seems to be a mythos invented in and ingrained in the fabric of the American psyche.
I think you're misinterpreting the parent commenter. They said that, in the U.S., speed limits are almost universally slower than the natural (i.e. prevailing) speed of traffic by 5-10 MPH, and so everyone driving at the prevailing speed of traffic is inevitably exceeding the posted speed limit by a bit.
That is definitely true, at least if you're not in the rightmost lane of traffic sandwiched between semis and box trucks. 75-80 MPH in 70 MPH zones is common, and it's not unusual to see people driving 85+ MPH.
> That is definitely true, at least if you're not in the rightmost lane of traffic sandwiched between semis and box trucks. 75-80 MPH in 70 MPH zones is common, and it's not unusual to see people driving 85+ MPH.
As an observation I'd agree. However, my impression of the parent commenter was that they implied that 5-10mph over the speed limit is the correct way to drive. Were that true, why wouldn't we just set the limit anyways? How did that limit get chosen?
Again, this goes back to the 85th percentile - the ways in which we decide how fast to go on a road or street are wrong. Of course, there's two things going on in this thread:
- Using tech to warn / limit drivers from being able to speed over an agreed upon limit.
- Engineering our roads so that the intended speed matches the prevailing speed of traffic.
Now I'm more in the second camp on this one than the first, I'll admit, but I don't think that we should just sit by and accept "some people speed, therefore we should all go faster" as a given. I have definitely witnessed people going 85+ MPH, especially if you're on an interstate far away from any city. However, this is not the norm and not really where the tech in TFA is going to be deployed.
Either way, the issue I take is with moralizing going faster than the speed limit. You might say it's the flow of traffic, but that just smells like justifying something that most drivers already want to do (go faster), instead of understanding the action (speeding) and its effect on the environment around us (increased danger, increased fuel consumption, increased noise pollution, etc).
Things must be crashing and burning over there in California with all the driverless cars.
Or you know, cars will just go the speed limit, 10 years from now and you won't even remember your brain simulated a shitty PID controller to hit some speed number subject to your irrational whims.
They won't. The rich won't let you dictate how fast they can go, so ferraris will be exempt. The upper classes will 'illegaly' turn off the limiter on their teslas 250d, but no sane sheriff will dare to enforce the law here. Middle class will be loosely complaint, the same way they observe the speed limit laws today. Only the lower classes will have no choice but to drive the 'chopper cars'.
There are still tons of cars from the 90's and early 2000's on the road (in particular the '01 Camry), it'll be at least 20-30 years and anyone with a new car being limited is just going to be harassed for driving way too slow relative to traffic.
And even once this is over, I think it's pretty often the speed limits and not the people driving over them that are in the wrong. I don't think slowing everyone down is a useful goal unless we adjust the limits up first.
I agree, but I don't think that would be a good thing.
I suppose if the end result was that speed limits got adjusted up 5-10 mph to match the speed of the road it would be fine. I doubt this would happen in general, and I know it wouldn't happen universally.
That might be right, but suddenly enforcing a limit when overtaking can easily cause an accident.
And speeding might not be that bad, many of them seem outdated or plain wrong, to the point where it's safe to go 10-20km/h faster, and in some places ridiculous not to go faster.
Come to Houston. You better be speeding on the highway or you're in danger. When everyone is speeding, and you're doing 60mph, you're a danger to yourself and everyone around you.
I would add another thing that you suddenly have a less clear agenda or daily goals to achieve.
Working as a IC you often have a backlog of work provided by someone else where it is their job to prioritise and structure that work for you. Moving into a management role it becomes your job to find and prioritise your own tasks.
It is very easy to feel like you aren't contributing or completing productive work as your workload and goals are now completely self defined.
Personally I have found it to be quite accurate. The smaller the company you work for the likelihood of 'flexibility' in your role increases. Those companies just don't have the headcount to have a full suite of engineers so rely on people dipping their toes in other tasks to keep the machine ticking along (e.g. frontend devs handling deployment).
With a larger company you will typically find they have already hired specialists to handle very specific tasks. You can always do some things but more often than not the rigor of corporate structure says "If you need anything done in dev ops, please speak to _Bob_ and he will sort it out".
Jumping from the challenge of constantly adapting to different tasks to being there to only do a single 'role' can be quite jarring.
This is something that has made me a bit nervous to switch jobs. I don't live on the coasts, so prospects are already a bit more limited. I've spent the past decade more or less on a team of ~5 engineers as part of a ~10 person company.
I did interview at a different startup a couple years ago, which was I think around 100 employees at that point. They claimed to be very lean and quick on their feet, but I got the feeling that we were talking about different levels of volatility.
I suppose the thing that actually keeps me where I am is I have both a very long leash and a high degree of influence. Hard to level-set that against other positions I see listed.
Code Heroes are a small mobile app company based in Brisbane focusing on mobile application development for medium to large companies. We are currently actively looking for experienced developers that have hands on experience with Flutter (Dart), or those with mobile app development history and a willingness to learn.
The job is full-time and onsite with a 6 hour workday. We are located the CBD of Brisbane.
With the speed this can generate its solutions, you could have it loop through attempting the solution, feeding itself the output (including any errors found), and going again until it builds the "correct" solution.