will turn into our digital assistants, serving to us navigate the complexities of the fashionable world. They may make our lives simpler and extra environment friendly.” Inspiring and fully unbiased assertion from somebody who already invested billions on this new expertise.
The hype is actual for AI brokers, and billions are pouring in to construct fashions that can make us extra productive and extra artistic. Laborious to disagree after I fortunately take pleasure in my morning espresso whereas Cursor is coding my unit checks. But, asking individuals in my community how they use AI of their day-to-day, their solutions usually point out anecdotal use circumstances, anyplace from “I exploit it to inform bedtime tales to my son” (I suppose that will not even be a use case for those who had extra creativeness) to “I exploit it to optimize my schedule” (Movement AI, please cease focusing on me for the love of god).
As a Information Scientist, my thoughts goes forwards and backwards between two conclusions. The FOMO a part of me that doesn’t wish to be late to the Robotic revolution get together, and the cynical one which thinks that there’s nonetheless an extended option to go earlier than synthetic intelligence truly turns into clever. To seek out out which aspect of my schizophrenic persona I ought to wager on, I’m going to make use of a easy but highly effective framework: reviewing all of the tasks I’ve labored on for the reason that starting of my profession and assessing how 2025 state-of-the-art AI fashions may have helped.
As we speak, we return to 2018. I’m a candid summer time intern at one of the crucial disruptive startups in America: Lease the Runway.
What the Mission was about
The Lease the Runway success heart in Secaucus, NJ, was the most important dry cleansing facility in america.
Within the Summer time 2018, as an Operations Analyst intern, I used to be given a fairly onerous drawback to consider: on a regular basis, the success heart was receiving hundreds of models again from throughout the nation. All of the gadgets needed to be first inspected, then would undergo a radical cleansing course of, earlier than being dried or receiving some particular therapies. This may very well be:
- Recognizing if the garment was stained in the course of the rental
- Urgent if it was too wrinkled and needed to be ironed
- Repairing if it had been broken
Most of those duties have been accomplished manually by completely different departments, and required specialised staff to be accessible as quickly as the primary batch of models have been reaching their division. Having the ability to predict days forward what quantity of models must be processed (and when) was essential for the success heart planning squad, in an effort to make it possible for each operations group could be staffed appropriately.
The complexity of the move made it even trickier. It was not solely about predicting the inbound quantity, but additionally assessing what a part of this inbound quantity would require particular therapies, the place and when bottlenecks may seem, and understanding how the work accomplished at one division would impression the opposite departments.
The 2018 Resolution
At this level you might surprise: given the complexity and the stakes of the venture, why was it within the palms of a younger inexperienced intern? To be truthful, throughout my 10-week summer time internship, I solely scratched the floor and wrote an insanely sophisticated Pyomo script that was later refined by a extra senior Information Scientist, who spent two years on this venture alone.
However as you possibly can think about, the answer was this big optimization mannequin taking as an enter the inbound quantity forecast for day-after-day of the week, the common UPH (models per hour, i.e the variety of models that may be processed in an hour) at every division, and a few assumptions on the proportions of models that will require particular therapies. The primary constraints have been on the timing and regularity of the shifts, and the variety of full time contracts. The mannequin would then output an optimized labor planning for the week.
How AI may have helped
Let’s re-clarify issues first: you’ll not see phrases like “AI-enthusiast” or “LLM believer” in my LinkedIn bio. I’m fairly skeptical that AI will magically clear up all our issues, however I’m occupied with seeing if with at this time’s expertise, one other method could be potential.
As a result of our method was, you possibly can say, fairly old fashioned, and required months and months of refinements and testing.
The primary restrict is the static side of the answer. If one thing sudden occurs in the course of the week (e.g a snow storm that paralyzes the logistics in some components of the nation, delaying among the inbound quantity), plenty of assumptions of the mannequin must be modified, and its outcomes have gotten out of date.
It is a answer that requires information scientists to go deep into the weeds, as an alternative of counting on an out-of-the-box framework, to depend on plenty of assumptions and to spend time sustaining and updating these assumptions.
Might AI give you a totally completely different method? No.
For this specific drawback, you clearly want an optimization mannequin, and I’m but to examine an LLM having the ability to deal with a mannequin with such complexity. One may suggest a framework with an AI agent performing as a Basic Supervisor, and counting on sub-agents to deal with the planning of every division. However that framework would nonetheless require brokers to have instruments that permit them to resolve a posh optimization mannequin, and the sub-agents would wish to speak because the state of affairs of 1 division can have an effect on all of the others.
Might AI considerably improve the “human-generated” answer? Attainable.
It’s at this level fairly apparent to me that LLMs wouldn’t make the issue trivial, however they may assist enhance the answer in a number of areas:
- Initially, they may assist with reporting and choice making. The output of the optimization mannequin may need a enterprise sense, however making a choice out of it could be onerous for somebody with no robust understanding of linear programming. An LLM may assist interpret the outcomes and recommend concrete enterprise choices.
- Secondly, an LLM may assist react quicker to sure sudden conditions. It may for instance summarize info on occasions that would have an effect on the Operations, comparable to unhealthy climate in some components of the nation or different points with suppliers, and as such, advocate when to rerun the planning mannequin. That’s assuming it has entry to good high quality information about these exterior occasions.
- Lastly, it’s potential AI may have additionally helped with making actual time changes to the planning. As an illustration, it’s sometimes predictable primarily based on the garment traits whether or not they would require particular care (e.g a cotton shirt will all the time must be ironed manually). Having a VLM scanning each garment on the receiving station may assist downstream departments perceive how a lot quantity they need to count on hours upfront.
Might AI allow Information Scientists to keep up and replace the mannequin? Sure!
It’s actually onerous to disclaim that with instruments like Copilot or Cursor coding and sustaining this mannequin would have been simpler. I might not have blindly requested Claude to code each constraint of the Linear Program from scratch, however with AI code editors being smarter than ever, modifying and testing particular constraints (and catching human errors!) could be simpler.
My conclusion is that an LLM in 2018 wouldn’t have trivialized the venture, though it may have enhanced the ultimate answer. However it isn’t inconceivable to imagine that a couple of years (months?) from now, brokers with enhanced reasoning capabilities will probably be refined sufficient to begin cracking these kinds of issues. Within the meantime, whereas AI may pace up mannequin iterations and changes, the human judgment on the core stays irreplaceable. This serves as a worthwhile reminder that being a Information Scientist isn’t nearly fixing mathematical or pc science issues—it’s about designing sensible options that meet evolving, usually ambiguous and never so nicely outlined real-world constraints.
Article 100% human generated
