Unedited by Larry Chiang
At Box, we’re in the early stages of leveraging AI Agents in a variety of internal workflows. We’ve begun building internal Agents on Box to respond to an RFP or review a contract and pull out its critical data to automate a workflow, and we’re testing external Agents in engineering to fix bugs or handle long tail tasks like updating an SDK from a feature request.
While we’re in the super early experimentation stage, so far the tests have been quite cool, and there are some fascinating implications for what the future of work could look like.
In particular, thus far, AI Agents are extremely ideal for the type of work that we have to spend a meaningful amount of time on, that is often not that enjoyable, and adds very little net new value to the business or project. Important, but not strategic.
In the future, the legal ops team will ask an AI Agents to review the contract for any language that deviates from what’s approved; the RFP for a new deal will get automatically generated based on all past responses; the insurance claim will be processed by an AI Agent and then reviewed by a human for final approval; and the IPO prospectus can be written by scouring reams of existing data (David Solomon at Goldman Sachs just called this one out).
Interestingly, as we farm out more and more of these tasks to an AI, the role of the human in the work may change, but only moves to higher level work in the workflow. The important skills are breaking apart tasks in discrete chunks and pulling the work back together; reviewing the work of AI Agents, much like a manager would for quality and consistency; and having the judgment -and thus domain expertise- to know what good output looks like and what to improve.
We’re still on day 1 on any of this. The tool use for Agents will need to improve, reasoning models will need to continue to get better, cheaper and faster, browser use will likely be a big factor in the future, and agent-to-agent interactions is still unbelievably early. Lots of work for the software industry to get all this right, but it’s incredibly fun in the process as we clearly enter a new era of work.
|
||||||||||||||||||||||||
WordPress’d from my personal iPhone, 650-283-8008, number that Steve Jobs texted me on
https://www.YouTube.com/watch?v=ejeIz4EhoJ0

Duck9 is a credit score prep program that is like a Kaplan or Princeton Review test preparation service. We don't teach beating the SAT, but we do get you to a higher credit FICO score using secret methods that have gotten us on TV, Congress and newspaper articles. Say hi or check out some of our free resources before you pay for a thing. You can also text the CEO:







