Surely this is a bearish development for frontier labs? Basically everyone expect for them is incentivised to see LLMs become a commoditised layer of the stack.
Once use cases stabilise, optimisation is immediate next focus.
On a related note, I foresee a lot of orgs (particularly 2nd/3rd tier quality companies) who took out company wide Claude code licences last year in order to “do something” with ai being very receptive to (a) the routing pitch and (b) the argument that 1st party harnesses aren’t incentivised to minimise token costs.
BTW, having just tested it - claude tag is another good example of the problem these labs have.
Interesting concept and I can see the power, however the idea that we would lock in dependency on a 1st party harness like this, which is SUPER token hungry if used in the intended way, is just madness.
X is full of VC backed businesses pushing token maxing, however for the majority of the economy it's not viable to adopt tools and use cases which can add variable costs of £50-£300 a month / employee with no clear ROI and no control over spend levers apart from binary on/off.
Surely this is a bearish development for frontier labs? Basically everyone expect for them is incentivised to see LLMs become a commoditised layer of the stack.
Once use cases stabilise, optimisation is immediate next focus.
On a related note, I foresee a lot of orgs (particularly 2nd/3rd tier quality companies) who took out company wide Claude code licences last year in order to “do something” with ai being very receptive to (a) the routing pitch and (b) the argument that 1st party harnesses aren’t incentivised to minimise token costs.
it's a debate I've been having a lot.. for the frontier labs, the TAM for frontier intelligence is getting smaller and smaller.
BTW, having just tested it - claude tag is another good example of the problem these labs have.
Interesting concept and I can see the power, however the idea that we would lock in dependency on a 1st party harness like this, which is SUPER token hungry if used in the intended way, is just madness.
X is full of VC backed businesses pushing token maxing, however for the majority of the economy it's not viable to adopt tools and use cases which can add variable costs of £50-£300 a month / employee with no clear ROI and no control over spend levers apart from binary on/off.
They are out of touch.