Aaron Levie
@levie
Followers
3M
Following
94K
Media
1K
Statuses
11K
ceo @box - unleash the power of your content with AI
Bay Area
Joined March 2007
AI agents with full tool use is going to be quite insane. Here's Claude using the Box and Linear MCP servers to take product roadmap docs from Box and turns them into issues to track. This is a small example of what the future of AI agent interoperability looks like.
22
23
224
The Box remote MCP server is now live! Now AI agents can all pull from the same enterprise content, instead of duplicating data across systems. AI agents using external tools and data is the future, and it’s going to be wild. You can go try it out in Claude right now.
The Box MCP server is now GA + live in @AnthropicAI's Claude. Securely connect AI agents to your Box content and give them the business context they need to be effective and helpful at work. Built on open standards, putting your data to work across your AI ecosystem. Read more to
12
19
118
We’re proud to welcome Eileen Gu as TCL’s Global Brand Ambassador. Together with the Olympic champion freestyle skier, we’re on a mission to change the world through passion and courage. Witness each moment of aspiration and triumph at the Olympic Winter Games Milano Cortina
10
7
92
The easiest ways to miss the full power of AI agents right now is to just not do enough with them. Most of the time people are just not pushing them far enough, and so are likely only benefiting from a small portion of the potential. It’s mind blowing when you talk to startups.
simply, the best vibecoding advice i can give is being more ambitious. if you're requesting a 10-min PR, ask yourself, could this model handle a 3-hour PR? . the answer is, way more often than you'd think: yes.
34
54
362
The paradigm of AI subagents is going to be super interesting. There was probably some hope or belief that a universal agent would be able to handle everything you needed in a workflow by stuffing all the relevant context into the context window. But even with larger context.
57
75
433
The context for AI agents to be effective is living inside of our workflows and enterprise data. This will cause existing platforms, that adapt quickly, to become even more useful, and will cause all new agent platforms to emerge in spaces that never had software before.
SaaS will be supercharged by AI Agents. @levie tells @MatthewBerman why domain-rich platforms are ideal terrain for specialized agents steeped in deep workflow context.
21
90
251
Almost every study shows doctors with AI perform better than those without. Now AI is achieving perfect scores in medical licensing exams. You will simply expect every professional services provider you go to will use AI in the future or you won’t trust the advice.
The era of superintelligence is here. Didn't predict the medical field would be first. Amazing work by team @EvidenceOpen in scoring a perfect 100% on the US Medical Licensing Exam
52
96
561
One subtle reason why the market size for AI Agents is going to be bigger than we realize is there are many use-cases where running multiple agents in parallel to solve the same problem is valuable. This was effectively never possible with most knowledge work before. It just.
65
129
710
AI Agents are a real transformation in software monetization. Traditionally, software was largely capped at ~$10-50 or so per month per seat for any particular software. There’s outlier vertical software, but it generally always had roughly this ceiling. AI Agents on the other.
We’ve gone so quickly from “not sure I want to pay $20/month for another AI coding tool when I already pay $20/mo for one” to “my $200/month subscription keeps running out of limits- help!!”. Devs actively using LLMs for work are trending to pay easily $1,000+/month soon….
39
60
369
Other than a few categories, most spaces are still wide open in AI. This will not be true forever because data and workflow moats will start to build up, but it certainly is temporarily true right now. AI has progressed so much in the past year that the models have solved a ton.
I sometimes hear prospective founders say they’re “too late” to AI . But because the models are improving so fast, in many categories / product types you’re advantaged by starting later . You don’t have to spend the time (and $) building infra/models that will be API available.
8
85
243
In a market full of competition, we claimed victory. And in July, we didn’t just survive. We dominated .🧵.
52
42
85
AI will blur the lines between many functions over time because you can now begin to do things either higher or lower in the stack, or expand to other adjacent functions. A very obvious area is that PMs should almost always be showing up with functional prototypes.
We are adding a coding section to all of our Product Managers interviews at @Shopify. We'll start with APM interviews. We expect candidates to build a prototype of the product they suggested in the case interview. There is no excuse for PMs not building prototypes.
38
104
358
At Box, we spend a lot of time testing Box AI with new models on unstructured data to see what they perform well at in real areas of knowledge work. As we've seen from the benchmarks, GPT-5 offers a meaningful jump in capability over GPT-4.1 in reasoning, math, logic, coding,.
23
81
271
It's clear that we're on a trajectory right now of AI models continuing to improve in capability across math, reasoning, logic, tool calling, and various domain-specific tasks that will get better as more training data continues to get generated. While there will be debates.
30
63
380
The paradigm of background agents, like we've seen in AI coding, will start to arrive in all areas of knowledge work. Most areas of work have similar dynamics as AI coding agents, where it makes sense to send off a request, check in on progress or get pinged with.
When will we be able to add AI coworkers to Slack to help take on misc operational work? . Updating spreadsheets, creating channels, running "cron job" business workflows, conducting Deep Research/Operator-style tasks. Same API as a human (bidirectional DM's, calls, shared links).
21
39
311
AI coding will just get better and better because it’s a massive market and all the AI labs, and many startups, know the space well and have an incentive to make their own productivity better. So they will just build the features they also need themselves. Virtuous flywheel.
Claude Code keeps on adding features that I didn't realize I needed - being able to start a dev server in the background and occasionally dip into the logs is super useful.
21
25
250
Core vs. context is a critical concept to think through when figuring out what people will rebuild themselves with AI. Companies bring in “core” functions that differentiate them. This is what their core product or service is, how they sell to customers, things that drive their
I think the risk that companies build their own systems of record - ERP, ITSM, CRM etc - is incredibly low. Companies are not stupid. They have no competence here, the stakes are massively high, and regardless of how easy it is, they would still have maintain and optimize it,.
31
66
588
Great thread. Whatever an AI agent is capable of doing, it can also be tricked into doing. You should assume if an agent can access data, that a user can eventually get that data too. Agent security, access controls, and deterministic guard rails will be critical.
we hijacked microsoft's copilot studio agents and got them to spill out their private knowledge, reveal their tools and let us use them to dump full crm records. these are autonomous agents. no human in the loop. #DEFCON #BHUSA @tamirishaysh
17
78
477
I think there are lots of examples where generated software and UI will work on demand, in particular in a personal context when the stakes are low and it’s just a byproduct of a chat system’s response. But I would have different expectations in an enterprise context. The point.
you have to be blind at this point to not see the future. the era of traditional software, as a stable object with version numbers, roadmaps & feature sets, is pretty much gone. we’re prolly just slow to admit it. future systems will be fluid, ambient, entangled with context.
56
60
443
Lots of conversation on what the future of software looks like in the enterprise. Here’s how I think it plays out. For deterministic workflows where the cost of getting something wrong is high, enterprises will have a tendency to pick core platforms for their most common,.
31
97
701
RT @mostlysports: Josh Jung hitting a walkoff homerun in extra innings to win the game was ELECTRIC @SportClips
0
5
0
While fascinating, the idea of AI generating every UI on the fly is probably less likely than people think. The benefits of hyper customization likely won’t outweigh having to re-learn an app each time you use it or the risks of things breaking in unexpected ways.
Anyone that has lived through a major redesign knows generating UI on the fly won't be a thing for most products.
98
62
814
It's sometimes hard to grasp the significance of the reasoning and logic updates that are starting to emerge in powerful models, like GPT-5. Here's a *very simple* example of how powerful these models are getting. I took a recent NVIDIA earnings call transcript document that.
52
101
829