Aaron Goldsmid
Related people
Episodes
Notes
The Product Podcast · Deel Head of Product on How AI Agents Do the Work of 600 Employees…
-
What Part of a Job Should an AI Agent Replace?
Aaron Goldsmid says an agent should replace the worst part of a job, the chores that add no value and still have to get done, not the job itself. Taking a three-year strategy from an agent and treating it as final would be a terrible idea. The agent can help draft it. The work to hand over is small judgment: approve an ordinary expense, or allow a hotel that is 10 cents over the limit because a night in New York and a night in Lisbon are not the same price.
Getting someone paid, calculating tax, or filing a visa still follows the same steps it has for about 100 years. Gather the information, fill out the forms, submit them, and wait for the result. Easy automation already moved a file or ma
-
How Does Deel Grow and Stay Profitable?
Aaron Goldsmid says Deel grows and stays profitable by pairing real product market fit with unit economics done up front, then building beside a customer who is already buying.
Companies usually pick one side of a split: grow faster or become profitable. Deel asks for the unit economics even when a product is wildly inefficient right now, and whether the work will become efficient later. A large services arm and an operations arm sit next to the software. In software, the habit was belief that scale alone would make a product profitable. Deel always did the math. Starting with a customer makes profit hard to miss, because nothing is built in anticipation of a buyer. The team finds a custome
-
How Did Deel AI Agents Add $140 Million in ARR?
Aaron Goldsmid says the figure Carlos cites, over $140 million in ARR from AI optimizations, the equivalent of the work of 600 people now done by agents, came from picking up change off the floor. Writing automation used to be the last artisanal craft, so a long list of manual steps never got automated: moving data from one system to another, reformatting a spreadsheet every month, or logging into a tool with no API and clicking one button.
The second unlock was taking that list out of engineering's hands. Deel's product Akai follows how a person would train an intern. Come shadow the task, watch it done, try it, then have the result checked. Someone who is not an engineer can capture the w
-
How Does Deel Train Its Compliance Model?
Aaron Goldsmid says Deel trains small language models on a wiki built from the company's own compliance knowledge. On global work and compliance, that small model is at least as good as Sonnet and is quickly approaching Opus.
Deel has thousands of compliance experts across its markets. The knowledge base grew organically, then the team put it into a standard wiki based on Wikipedia so people could reach it easily and treat it as a source of truth. Language models are focused on the corpus, and small language models are now trained on the same knowledge base. The upside is second-level responses versus waiting five to 10 to 15 or 20 seconds, at a fraction of the cost.
The investment in the
-
How Does Deel Get Data Without Waiting for a Meeting?
Aaron Goldsmid looks after Deel's data organization, where semantic layers and small AI analysts answer a data question without a ping. Deel is globally remote, so the person who has the information is often asleep. The old loop waited until a meeting could be scheduled, which turned wildly inefficient, or until that person was on Slack. Asking the semantic layer short-circuits the wait, and people only need to be awake for the information to flow freely.
Deel has grown very quickly, and engineering entropy is inevitable. A core promise is an all-in-one platform with a better-together effect, like G Suite, where the pieces naturally work together. One end of the spectrum is everyone writing
-
How Do Two-week Reviews Keep Deel Product Teams Aligned?
Aaron Goldsmid says a business review every two weeks keeps product, engineering, and design from diverging for longer than two weeks. The group covers what happened in the last few weeks, what shipped, what is planned, how things are going, and where the blockers are. It is not a time for pretty presentations. Teams can bring as much dirty laundry as they have. They do not have to agree, but there can be no daylight between them on what the opinions are.
Ops and sales sometimes join line-level leadership, one level above, or him. Side conversations create the drift: product gets one set of advice while design and engineering hear the same point in slightly different words. At a certain sca