janstice

On the other hand, SpaceX has a handful of very large clients, and around 12 million small clients, which probably provide a less clumpy income.

If you were paying commercial token rates, what would the cost have looked like?

This the the pricing model for Anthropic now too for business customers over 150 seats - under 150 you can take the Team plan at $25/mth & it’s similar to the personal plans, but Enterprise you’re billed by the token (with a $20/seat minimum but the tokens are pooled together). This is the AI Uber moment where the VCs stop subsiding your fares.

The frameworks-and-tools make for good blog fodder too, as they are quite applicable across a range of areas, so many readers will find something that resonates with them, and claude-code-is-pretty-good-these-days is a less blogworthy topic.

Is your phone connected to some work mobile device management? I could imagine someone has a jinxed Jamf or intune rule that is pushing things out.

But when you build a skyscraper you don’t one shot a completed building that stays static its entire life - you build a set of empty floors that someone else designs & fits out, sometimes years after the building as a whole is commissioned, usually several times in the lifespan of the superstructure.

And in the fitting out there often are things that exist only to get customer feedback (of sales), such as model apartments, sample cubicle layouts etc.

So yes, you are right that engineering can guide us to building something right first time - the hard part from software perspective is usually building the right thing, no the thing right.

An interesting analogy I came across once but could never find again is that with software systems, we’re not building a building, we’re designing a factory that produces an output - the example was a mattress factory that took in raw rubber feedstock & cloth and produced mattresses.

At least AI is (and unlike many contract dev shops) keen to write unit tests…

Workday’s student offering is designed as a full student management offering like Banner et al, with the carrot that it’s internally integrated into the financial & HR systems, which avoids another vendor and also a massive and ongoing finger pointing exercise.

It’s also one of the few from-scratch cloud-first student management solutions.

Your system was configured by muppets if you don’t have a search box - it’s a massive beast that like all enterprise-grade software is a toolbox for you to bend to your will, but the downside is that if your configuration people don’t have empathy for the users (and looking at you especially, contract architects) you end up with a system that is optimised for whoever talks with the vendor, and not for anyone else.

Try this one: https://www.theguardian.com/world/2025/nov/11/milan-prosecut...

To be fair, if I paid $30k+ for an H200, I’d want it to be making money 24/7 rather than idling, so the idle power draw would be strictly theoretical.

A two week delay on including new versions would probably work more or less as well with a bunch less effort, but a local proxy looks like it’s going to be a lot more common very soon I’m guessing.

I think the reason that the big guys don’t make money out of old films is that if they did they’d be on the hook to pay the cast & crew(‘s retirement plans).

And I guess if there’s a whole-of-life policy sold in 1962 that hasn’t been terminated - I guess there’s a lot of grandfathered rules that are just easier to keep in their original systems.

I suspect the product telemetry would be more useful - things like success of interaction vs requiring subsequent editing, success from tool use, success from context & prompt tuning parameters would be for valuable to the product than just feeding more bits into the core model.

Oddly enough this also caused similar issues in classical orchestras - in the 90s a bunch of top flight Eastern European and Russian musicians raised the bar of orchestras in places like NZ, with the side effect of having fewer seats for younger musicians to move into.

Or even Docker Desktop, which is a bunch of $$ that no-one expects (and I think Microsoft is still recommending in the WSL docs).

For example, if someone installs the wrong version of Oracle Java on a VM in our farm, the licencing cost is seven figures as they want to charge per core that it could conceivably run on - this would be career-limiting for a number of people at once.

Sure is - and o3 is missing from the OpenAI models that Azure is serving, which I suspect isn’t a coincidence - if OpenAI has some secret sauce that lets them undercut resellers this might shake up agreements for a bit.

Or rather; the candidate didn’t have a visa, the employer would need to jump through a bunch of hoops, then a multi-month wait, OR can just hire one of the many on-shore candidates available on the market.

I think previously these sorts of offshore people were picked up by big bodyshop contractors, who could reliably place someone (and afford to have someone on the bench for a few weeks if needed) - since a massive bunch of government contracts were cancelled over the last few years this mode has dried up.

In that case, any suggestions if the answer was looking for workflow engines? Ideally something that will work for no-person-in-the-middle workloads in the tens of seconds range as well as person-making-a-decision workflows that can live for anywhere between minutes and months?

Management consulting - I expect less than 20% of what a random 24 year old in a suit that you pay $3000 per day produces is actually specific to your business problem, and the rest is formulaic.

I’m even happy to listen to generative music, so long as it’s orchestrated (haha) by musicians using musical taste to make musical decisions, rather than a pastiche of the worst derivative house you’ve ever heard by a rando with no intent.

It’s harder to demonstrate growth & development in the same job for 7 years - if you have a couple of job changes it makes a more natural narrative of how you are professionally developing - not impossible at a single org, but needs a bit more of a story of projects you delivered and how you are not the same person as 7 years ago.

The two fast.ai courses end up implementing Stable Diffusion from scratch if that’s the sort of thing you are looking for.

I wonder if it’s a tracking off switch or a panic button? I used to work for a fleet tracking SaaS, and some customers with unionised workforces needed a way to disable tracking, and panic buttons were common too (although less so in Europe).

Also shared tablets don’t really work as kids homework devices.

And when you set up 1Password, make sure you also get the CLI going, so passwords & shared gunk that's needed to access other people's services can be scripted, and when the passwords, etc get rotated no-one needs to know because no-one needs to store them.

In 60s/70s France 14 year olds could ride motorised bicycles, so Vespa sold a Pedalo - a regular (but lower-powered I think) Vespa scooter with pedals as well as 2-stroke engine - you had to put the engine in neutral to pedal (I bet not many 14 year olds did this).

SOC2 stuff (which we all know is largely theater)

SOC2 is only theatre if you (a) you already have good practice, and (b) can demonstrate that you have good practice. If your practice isn't good enough (like the whole notion of security controls is a foreign concept), and sure there's a lot of boilerplate to work through -- but the whole point of a SOC2 Type 2 report is that you only have to demonstrate once to the auditor, rather than to each customer each time.

Having to get internal security sign-off for a non-audited SaaS vendor -- really, life's just too short for that most of the time, and if there's choice of two more or less equivalent providers we go with the certified one every time.