Investors are pricing in a 32.6% AI productivity boost for software engineers Investors appear to be betting that AI will deliver substantial gains in software engineering productivity, according to economists who used stock market movements to estimate the technology's expected impact. Economists affiliated with the University of California, Berkeley (UCB) and the London School of Economics and Political …
Do you think any of these self-styled ""economists" bothered to look at historical events or did they just plot out the most naive graph possible (taking care not to dribble on the page) then plonk the ruler down about halfway through the dots and draw a straight line off the edge of the paper?
Remind us again, what do stocks do when in an unsustainable bubble?
Reality is going to hit some very hard.
Firstly we have the few real-world studies which show developers taking 20% longer to do stuff via AI, although we could (rightly) ask whether the results achieved are better than would have been achieved otherwise.
Secondly, we shouldn’t forget the AI subscription and usage costs, which are highly likely to exceed the cost savings, hence we are down to the price executives are willing to pay for the difficult to define market advantage such a productivity boost might bring.
So I expect in a few years there will be companies with a smaller designated IT workforce but with substantially higher IT OPEX and thus thinner profit before tax. Naturally, they will all complain the cause of their woes is the salaries and taxes they are having to pay…
From talking to more senior developers (and real developers, not make-believe-ones like I am): They can safely use LLMs because they can identify BS output. It does speed up some things. Not all. And not all developers have the skill and experience to do the same. Plus they actually enjoy programming and problem solving, so they use LLMs for the "boring" stuff.
Yeah, that's anecdotal evidence et c.
Current workplace so far has only an on-prem LLM that sucks. A lot. Due to regulations we cannot use anything off-prem.
And, in an era where AI slop is overwhelming the scientific publication industry, that could very well be the point.
There's no doubt that AI is making some experienced developers considerably more productive. But this is not only at the expense of less experienced ones, who are unlikely to get a look in, I'd also suggest it's far from permanent. Once this becomes standard, then competition will start to drive down prices.
> a permanent 32.6 percent productivity increase
... it is only to show they have a sense of humour.
I wonder just how wide the (never reported) error bars are on this prediction.
I would offer the following:
1.) Things are never as bad as they appear
2.) things are never as good as they are made out to be
3.) all the unknown factors add at least a +/- 100% level of uncertainty
4) timescales are even more uncertain
5) predictions (as observed by an American sports personality) are tough, especially about the future
The reported economists are stating the figure they ascertained that investors have (in aggregate) decided that AI/LLMs will give them. (So decimal point accuracy is purely the result of that calculation - not a forecast). Not that they are themselves predicting such productivity savings.
Also, by confusing this a significant point is being missed:Investors will expect to see those savings. Indeed will likely demand them- and base their valuations on them.
The consequences of which could be significant, whatever happens. Not the least that companies' C-Suites will be ordered in no uncertain terms to provide these cost cuttings.
Whether it's based on reality is of no relevance. A prediction will have been made and factored in to the expectations. Jobs will be lost and productivity targets will potentially not be met
If planning quality and speed at management level improved 32.6% then there would be a productivity boost, but not necessarily due to AI.
The bottleneck in software engineering has always been in the planning and that remains so regardless of AI.
Spec driven development is touted as being the solution, but if management couldn’t be arsed to write any decent specs before AI was a thing, what makes anyone think they will now?
Use AI to generate specs from a single sentence?
Garbage in = garbage out.
As usual, at most companies, it’ll be software engineers painfully trying to tease out what the business actually wants = no time saved
Our Company Director recently told the IT Department that he needed them to upgrade the whole department to licenses with CoPilot and that he expected to see a 30% productivity increase. At least I now have a idea of where that figure might have possibly come. Now if only that also came with the flip-side of this paper's conclusion, that productivity increases are paired with increases in compensation...
>” Now if only that also came with the flip-side of this paper's conclusion, that productivity increases are paired with increases in compensation...”
Missed the small print, or the unwritten rule:
As it was your Director wanting the change, implicit in wherever they read the 30% productivity increase, the compensation increase was implicitly to the directors…
