Does a Speed-Up Even Help You?
A few folks joined me and my cofounder, Ehsan, to hear how relatively simple machine learning methods can give significant speedups in otherwise laborious computations.
You can access the recording here https://www.crowdcast.io/c/r3pdzydr76v0 if you want to catch up.
On a higher level, what I like about this sort of work is that it highlights the importance of thinking about your whole system: When implementing solutions to speed up costly processes, the question should always be: “And what does that lead to?” which is also crucial for assessing which workflows to speed up with AI Agents. If speeding up one part of your system just leads to a pileup of untouched work in another part, you don’t gain efficiency; you destroy it, because all that surplus now clogs up the proverbial pipes.
This is where you’ll want to look at the overall flow of work: Where do things get stuck? Which parts of the system are choking and which are starving? Even without AI, this is a critical analysis. Are your developers churning out massive amounts of code that then get stuck in a lengthy review process? Pushing the devs to produce even more code, faster, won’t do you any good then. Optimize not for the speed of an individual stage in the pipeline. Instead, optimize for the overall throughput: From the start of a task to its completion, where does it spend the most time?
And don’t neglect the “interaction” of separate work streams, either: If part of producing value in your company depends on specialists using their specialist skill, you can either try to get them to apply that skill faster, or you can free them up to do more of that special skill by empowering them to do less of another thing. In a concrete example, if you run an award-winning restaurant, the way to serve more diners, faster, isn’t to exhort your star chef to work faster. It’s to get someone else to clean their dishes and chop their ingredients for them.
That’s where I’m confident AI will unlock more value, at least in the short term: by allowing specialists to spend more time on high-value tasks instead of low-value administrative overhead.
