Oliver Andersen

Currently building @ai_erp. Previous Head of Product at Uber, DeliveryHero

[Oliver Andersen @ofandersen]

if production scheduling still lives in a spreadsheet after the erp cutover, then the work center stays idle while someone rebuilds the week by hand.

[Naologic @ai_erp]

You still rebuild the week in Excel when you leave a legacy ERP.
Monday 07:30. Your week is still a spreadsheet.
The down press stays idle while you type the plan.

[Oliver Andersen @ofandersen]

n the work center stays idle while someone rebuilds the week by hand.

[™³k @_tm3k]

A retard in motion is always better than a genius at rest

[Oliver Andersen @ofandersen]

Agreed that AGI isn't here yet, but if you view your daily work in various clusters (fx for me as a founder: coding, design, research, marketing, sales, BD, hiring and GTM), then you can build out a capable network of agents that can alleviate you of a considerable amount of work that'd otherwise fall on a human.

A lot of tasks fall in the category of "I'll get to them next week" (and next week continues to always be exactly one week away!), not because they aren't important, but because they aren't P0 or P1 tasks.

And again, if the present state is the worst agentic work will ever be, then I'm excited to see what's ahead.

with all that said though, my post was actually a point about improved loops and prompting of the agents (similar to the effort you'd invest to correcting a human whose approach needs adjustment).

[Oliver Andersen @ofandersen]

@VedantOps the board says available. no material and a late setup still means the machine is waiting.

[Oliver Andersen @ofandersen]

Fantastic example of a truly HORRIBLE product: Twilio.

A useful marker for how solid your product is: People shouldn't want to jump out the window after interacting with it.

[Oliver Andersen @ofandersen]

If you can’t play padel, you play pickleball

[Bautista Bassi @BauBassi]

hoy se me sentó un italiano adelante con esta remera

[Oliver Andersen @ofandersen]

@MrGadan the 08:00 hour is the pipeline. we lose operations directors the minute the first call is an AI deck.

[Oliver Andersen @ofandersen]

@houmanasefi 51 percent includes Copilot in the office. a model that is allowed to pick the next job on the line is a different number.

[Oliver Andersen @ofandersen]

@nathanbenaich @odysseyml @raais we're seeing this at @ai_erp. operator preference on which job to run next beats another month of scaling the planner.

[Oliver Andersen @ofandersen]

@igor years of stored data doesn't help if nothing writes back. at @ai_erp the record has to close the next job.

[Oliver Andersen @ofandersen]

👀

[Polymarket Money @PolymarketMoney]

BREAKING: Nvidia declares physical AI could become “10x larger than digital AI” with Jensen Huang predicting every industrial company will become a robotics company.

[Oliver Andersen @ofandersen]

@nikolai_erp we're seeing this at @ai_erp. the gantt is out of date by 10am. shop floor data has to pick the next job.

[Oliver Andersen @ofandersen]

@michaelaubry This resonated extremely hard

[Oliver Andersen @ofandersen]

@sahildarz agreed, often times I have to go poke around before the issues surface

[Oliver Andersen @ofandersen]

Autonomously running agentic workflows are interesting because they challenge me to accept (with the present state of AI) workers who do the job 90% as well as I’d have hoped, not 100%. The same challenge I’ve faced with human-reports throughout my career.

Now imagine how much time and you’re willing to invest on humans to correct behavior, educate and mentor - if that same willingness is transferred to an agentic workforce, that is likely the path to get from 90% task completion correctness to 100%.

[Oliver Andersen @ofandersen]

@PaulGiese17 vendor is the demo problem. the one that stops a close is item master. four plants, four names, one part.

[Oliver Andersen @ofandersen]

Customer stories we're seeing when moving folks off of their legacy systems

[Naologic @ai_erp]

Plants we migrate still walk the office for a paper job packet.
Monday 06:40. The packet is still on a desk.
The first press sits idle while someone walks the office.

[Oliver Andersen @ofandersen]

i did create a bunch of awesome agents with @bot, but really struggling to see how I'll be able to run these workflows and keep expanding them without blowing through several thousand every month just in grok bot cost (nevermind cursor or actual grok limits).