It seems inconceivable in 2026, but in the late 1960s my parents dropped me off for an unaccompanied ~10-hour train trip from Melbourne to Sydney. That trip included multiple stops at different stations along the way, plus a change of trains as we crossed the state boarder. As far as I'm aware, there was nobody on the train who was watching over me.
What are peoples' current suggestions for using Claude Code with a locally hosted LLM running on regular consumer hardware (for the sake of discussion, assume you're spending $US500-ish on a mini PC, which would get you a reasonably decent CPU, 32Gb RAM and a cheapish GPU)?
I get that it's not going to work as well as hosted/subscription services like Claude/Gemini/Codex/..., but sometimes those aren't an option
As someone who's been coding for several decades now (i.e. I'm old), I find the current generation of AI tools very ... freeing.
As an industry, we've been preaching the benefits of running lots of small experiments to see what works vs what doesn't, try out different approaches to implementing features, and so on. Pre-AI, lots of these ideas never got implemented because they'd take too much time for no definitive benefit.
You might spend hours thinking up cool/interesting ideas, but not have the time available to try them out.
Now, I can quickly kick off a coding agent to try out any hare-brained ideas I might come up with. The cost of doing so is very low (in terms of time and $$$), so I get to try out far more and weirder approaches than before when the costs were higher. If those ideas don't play out, fine, but I have a good enough success rate with left-field ideas to make it far more justifiable than before.
Also, it makes playing around with one-person projects a lot practical. Like most people with partner & kids, my down time is pretty precious, and tends to come in small chunks that are largely unplannable. For example, last night I spent 10 minutes waiting in a drive-through queue - that gave me about 8 minutes to kick off the next chunk of my one-person project development via my phone, review the results, then kick off the next chunk of development. Absolutely useful to me personally, whereas last year I would've simply sat there annoyed waiting to be serviced.
I know some people have an "outsourcing Lego" type mentality when it comes to AI coding - it's like buying a cool Lego kit, then watching someone else assemble it for you, removing 99% of the enjoyment in the process. I get that, but I prefer to think of it in terms of being able to achieve orders of magnitude more in the time I have available, at close to zero extra cost.
I work in a space where I get to build and optimise AI tools for my own and my team's use pretty much daily. As such I focus mainly on AI'ing the crap out of boring & time-consuming stuff that doesn't interest any of us any more, and luckily enough there's a whole lot of low hanging fruit in that space where AI is a genuine time, cost and sanity saver.
However any activity that requires directed conscious thought and decision making where the end state isn't clearly definable up front tends to be really difficult for AI. So much of that work relies on a level of intuition and knowledge that is very hard to explain to a layman - let alone eidetic idiots like most AIs.
One example is trying to get AI to identify security IT incidents in real time and take proactive action. Skilled practitioners can fairly easily use AI to detect anomalous events in near real time, but getting AI to take the next step to work out which combinations of "anomalous" activities equate to "likely security incident" is much harder. A reasonably competent human can usually do that relatively quickly, but often can't explain how they do it.
Working out what action is appropriate once the "likely security incident" has been identified is another task that a reasonably competent human can do, but where AIs are hopeless. In most cases, a competent human is WAAAY better at identifying a reasonable way forward based on insufficient knowledge. In those cases, a good decision made quickly is preferable to a perfect decision made slowly, and humans understand this fairly intuitively.
I started coding in the 70s, loved it then, still love it now and LOVING the emergence of Gen AI tools.
For perspective, the IT industry went through a similar change with the emergence of search engines ~30 years ago. At that time, a big part of the value of a software "expert" was in their ability to remember and recall lots of info (most of it of dubious value, to be fair). These experts usually had shelves of well-thumbed books on all sorts of topics, and could recall obscure info from these books seemingly at will. With the emergence of AskJeeves, AltaVista and eventually Google, suddenly nobody needed to remember anything OR even know where to find it - with a simple search, you could get nearly all the info you needed.
I can still remember the panicked response to this brutal change from the senior IT people I worked with at the time...
Did the demand for skilled developers dry up? No
Nor did it end with
- introduction of COBOL (designed so that non-coders could write code),
- PCs (surely leading to the end of systems programming as a career),
- spreadsheets (so accountants no longer needed programmers),
- 4GLs (designed to greatly simplify coding; report writing in particular),
- Visual BASIC (so the world would no longer need C programmers; anyone could learn to write BASIC),
- Microsoft SQL Server (nobody would need mainframe databases any more, so all those mainframe jobs would disappear)
- object oriented coding (all those code reuse possibilities! Very quickly programming should devolve to just glueing together other peoples' code),
- open source (because inevitably any tool of value would soon have a competitor that was free, destroying the value proposition of companies that wrote software to sell),
- Linux (how could Windows compete with free? Shed a tear for all those soon-to-be-unemployed Windows experts)
- NoSQL (because the need for "legacy" databases like Oracle, DB2, Postgres, MySQL etc. would surely go away)
- etc., etc., etc.
The reality is that you still need a grounding in software development to do coding well, even with AIs. I'm absolutely loving how quickly I can create solid code with the assistance of Gen AI - lots of tasks that used to take me a week I can now knock over in a few hours.
I also notice how many people are struggling with how to use Gen AI tools for coding tasks - my take is there's 2 distinct skills you need: knowledge of how to do software development well, and knowledge of how to use Gen AI tools for coding. Having the first doesn't automatically lead to the 2nd - you have to put in the time to learn about Gen AI, THEN work out how to fit Gen AI tools around your current workflow, THEN work out how to optimise the way you work with your new idiot savant buddy that has perfect recall.
That whole process (new tool appears -> learn about it -> work out how to fit it into my current workflow -> optimise my workflow) has basically been my entire career in a nutshell.
People have been predicting the demise of programmers for my entire career (40+ years now), and so far they've been wrong every time. For each new disruption that appears, the key has been to embrace it and adapt how you work accordingly.
Gen AI may indeed be different and kill off all programming careers overnight, but so far I'm not seeing it
> Most authors do not support a way to pay them directly.
I think this is the problem that should be addressed.
Musicians went through a similar process in reverse order: first Napster ("piracy") then streaming services (analogous to Kindle/Amazon, where a huge 3rd party inserts themselves between content creator & consumer). Eventually some musicians twigged that they were getting screwed every way, so they set up ways for fans to pay them directly or via a less money-hungry intermediary (e.g. Bandcamp).
Not a perfect solution by any means, but if book authors feel their situation is bad enough, they could look into how musicians are dealing with it.
I'm probably not alone in thinking I'd far rather pay an author directly than Amazon or book publishers.
While I've done more than enough Powerpoint presentations telling clients what they already knew but didn't want to say out loud, there are some circumstances where bringing in a consultancy is a very good option.
Some examples:
As a software/cloud/data/AI/cyber guy (I wore a few different hats over the years...), I regularly caught up with buddies working in legal, tax, audit, retail, space travel(!!) etc. for coffee chats. It's surprising how often those of us who specialise in one domain had breakthrough moments from offhand chats with specialists in other domains. Very few people get the opportunity to have these sorts of conversations, and it's amazing how often you learn something relevant for your own work situation over a quick coffee.
When I needed expertise in one of those domains into one of my projects, I could send a message and almost always get someone on a call within a few hours. Very few organisations could get e.g. a high-ranking ex-NASA official on a call quickly to pick their brains, but I could.
Lots of times organisations don't have the deep expertise and/or available people to deliver on their internal projects. When a major rail transport provider needs to work out how to going to deal with new government critical infrastructure regulations for their IT systems, it's consultancies who can pull all the right skills together to help them out.
When there's a critical shortage of available IT skills in the marketplace, companies use consultancies to top up their workforce. Here in Australia, there were nowhere near enough GCP experts to go around for the last 4-5 years, so companies could either try to hire the very few people around at exorbitant rates, or tap a consultancy for resources.
Big 4 consultancies in particular throw high-quality training at their technical staff like nowhere else I've seen. One reason: quality training = billable hours. I had people around me burn out from too much training, and I'm pretty sure regular companies don't have that problem. For all the pointless Powerpoint presentations we did, there's a sh1tload of technical expertise sitting in Big 4 consultancies, waiting to be tapped.
Companies are always struggling with how to use the latest IT shiny tools properly. Right now it's AI - how can I use it to save costs or increase productivity? What are the ethical and legal implications that come with AI, and how can we deal with them? How can we deploy AI solutions securely? Which of our business problems are the best fit for AI solutions? How do we train our ops staff to keep these AI solutions running? - the list goes on and on.
Now a lot of people here in HN know how to do these things, but how does a regular business tap into that expertise and filter out the bullshitters? The answer is they go to a consultancy that (they feel) they can hold accountable.
On that point, sometimes execs in a company simply need someone to shield them from blame for unpopular decisions like mass layoffs. It's pretty well know that consultancies do a lot of that type of work, and they (justifiably or otherwise) cop a lot of crap over it so some exec can say "I didn't do it".
I'm probably coming across as a huge fanboy of consultancies, but remember - I'm an ex Big 4 guy. I think their influence is too large, particularly in government policy making; I've encountered way more sociopaths in Big 4 leadership roles for it to be down to chance; by any measure staff turnover and burn out is very high, and I'm convinced that's by design.
> A Sony Walkman-style device that you can give to children so they can ask questions to an LLM. It should be voice-first, and focused on explaining things. There shouldn’t be a single screen on the device. Offline-first would be a plus.
My grandkids (5 and 3) spent about 2 minutes learning how to use it, then bombarded it with "tell me a story about a unicorn named Bob", "can dogs be friends with monkeys?" and so on. In every case it gave a reasonable answer within a few seconds.
I'll be amazed if these things don't wind up embedded inside toys by Xmas. When they do, I'll be in the queue to buy one
Right now, DeepSeek feels like it comes from the most trustworthy source, which is not something I would've said a few months ago.
In the short term I'll keep using OpenAI, Llama, Claude and Perplexity for what each does best. In the mid term, I'm looking for replacements for all 4
I was 7 years old