Dropbox VP’s Greatest AI Workflows for You to Take Again Time at Work

Pricey subscribers,

At present, I wish to share a brand new episode with Morgan Brown.

Morgan is the VP of Product for AI at Dropbox. He gave me an inside have a look at his high AI workflows, together with how he makes use of AI to anticipate exec questions. We additionally talked about constructing Sprint, Dropbox’s new AI agent for work, and the way AI will reshape PM.

Watch now on YouTube, Apple, and Spotify.

Morgan and I talked about:

• (00:00) How Morgan makes use of AI as a companion for every part he does

• (01:26) Morgan’s 3 finest AI workflows to save lots of time at work

• (06:07) Inside Sprint: Dropbox’s AI agent to search out something at work

• (14:35) The analysis course of behind Sprint’s search outcomes

• (18:49) Methods to ship quick whereas movement would not equal progress

• (22:09) Methods to stability inner alignment with buyer obsession

• (27:27) What Morgan realized from Drew (Dropbox), Tobi (Shopify), and Zuck

• (35:39) The three kinds of PMs Morgan seems to be for within the AI period

• (38:02) What product administration will appear like in a single yr

Welcome Morgan! To start out, what are your high 3 AI workflows to save lots of time at work?

I went by way of yesterday and realized there wasn’t a single factor the place I didn’t use AI as a companion. It is the primary tech because the iPhone that is been this pervasive.

My high three use circumstances are:

1. Get pre-read takeaways. I’ve a immediate that I take advantage of to evaluate paperwork. It seems to be for solutions to questions like: What is the job to be carried out? How do we all know this drawback exists? What are the assumptions?

2. Simulate govt questions. I’ve prompts for everybody I meet frequently, like Drew (Dropbox’s CEO) and our CTO. I run assembly agendas by way of these prompts to get questions they could ask. It is like getting a pre-brief.

3. Automate AI report takeaways. I’ve a ChatGPT script that runs every day, giving me the highest three AI papers from arXiv, summarized with highlights on that are probably the most groundbreaking to data work versus incremental.

Okay, let’s speak about Sprint, Dropbox’s AI agent for work. What does it do?

We constructed Sprint as a result of data employees spend 160 hours a yr simply searching for stuff and switching between contexts.

Sprint is an AI multimedia search engine that helps you discover and create content material quick. Whereas most AI search merchandise deal with textual content, Dropbox’s heritage is in artistic and advertising fields the place work consists of pictures, video, slides, PDFs, and extra.

Our key design philosophy is that:

Every time we are able to reply the query straight, we do not wish to ship you to a doc.

How can I collaborate with others utilizing Sprint?

Now we have a characteristic referred to as Stacks, that are collections of docs, hyperlinks, and supplies round any mission. I’ve stacks for my H2 roadmap, GTM efforts, and extra.

You’ll be able to add third events like advertising companies and everybody has entry to the identical context with far more related chats than common chat.

How do you run AI evaluations for Sprint?

We run a strong multi-tiered analysis course of:

1. Golden datasets: Created by our PMs and subject material specialists for particular workflows like product administration. An instance golden dataset may cowl: “I seek for a method doc or PRD, and the precise stuff reveals up within the outcomes.”

2. LLMs as judges: We consider the complete search outcomes web page for relevance, not simply particular person outcomes. LLMs fee each single hyperlink – whether or not it is pictures, movies, or paperwork – then we are able to see these rankings in our inner builds. This offers us suggestions to enhance the system.

3. Actual-time consumer suggestions: Since Sprint is used throughout Dropbox internally, we now have 1000’s of individuals utilizing it continually and giving us suggestions.

The latter is very vital. We be taught from how completely different groups use it from buyer assist to product administration. This offers us actual utilization patterns to know whether or not we’re nailing the search outcomes and chat classes. We use that information to enhance the educational loop over time.

Nice, I’d love to modify gears and canopy a few of your product administration sizzling takes subsequent. You’ve posted about how vital shifting quick is however have additionally mentioned that “movement would not equal progress.” What’s progress then?

Coming from biology, I feel by way of techniques. Each time you place one thing into an ecosystem, both one thing adjustments or nothing adjustments. Inside the “one thing adjustments” bucket, it both impacts issues positively or negatively.

I like transport as a result of it will get us out of hypothetical conditions and placing concepts into contact with actuality.

How do you align inner stakeholders to ship quicker, particularly in giant corporations?

I work backwards from the shortest path to fixing a consumer drawback.

The secret’s being specific about what you do not know:

1. What’s my speculation?

2. What proof do I’ve?

3. The place do I lack proof?

That is the place pace alternative comes from—fast analysis, buyer advisory board calls, mining Gong requires gross sales conversations, and extra.

For instance, what’s most vital to us—monetization, consumer satisfaction, process success, or authorized threat? These inputs are often implicit. I push my crew to state the ideas and rank their significance. Then when individuals debate options, you may debug: Is it an possibility drawback, precept drawback, or objective drawback?

I used to be speaking to Hiten Shah who works with you and he mentioned you had some nice recommendations on how founder-types can regulate to working at large corporations. Are you able to share these?

Positive, let me begin with a narrative first. In my first 6 months at Fb, my arguments simply weren’t touchdown and I used to be apprehensive I would not final.

My supervisor pulled me apart and mentioned: “You are relying an excessive amount of on private expertise and analogies from the previous. Individuals do not care what labored earlier than—they need confidence in what is going on to work now.”

That actually caught with me. Style and sample matching are critically vital, however they must be paired with driving alignment throughout individuals with completely different views.

Every time potential, it’s important to make your concept defensible and evidence-based. The place it is based mostly on style and is subjective, it’s best to name that out explicitly. Be actually clear on what you are attempting to resolve and the way.

It is like what we inform youngsters at school: present your work. Founders must take that internalized perception and externalize it in a manner that may be seen, challenged, and accepted.

What did you be taught from working with leaders at Meta, Shopify, and Dropbox?

1. Meta was nice at utilizing information to drive selections. Everybody used the identical language: Did this make retention higher? Engagement? Development? The choice-making framework is crystal clear which allowed the corporate to maneuver shortly even at extraordinary scale.

2. Shopify is data-informed however led by Tobi’s imaginative and prescient. Tobi would inform our development crew: “Do not run random experiments to maneuver metrics. Have a viewpoint on how this helps make extra entrepreneurs, then experiment from there.”

3. Drew (Dropbox CEO) desires to deeply perceive your rationale. He is some of the technical founders I’ve met. He thinks about merchandise not simply from the consumer facet, but in addition from technical functionality. Till you really perceive what’s technically potential, your possibility set is just too small.

Let’s wrap up by speaking about the way forward for product administration. First, what expertise do you search for when hiring PMs to work on AI merchandise?

I search for three issues:

1. Consolation with ambiguity. Can you are taking ambiguous conditions and give you a transparent method? With AI, issues change each 24 hours—you want to have the ability to body issues, put bounds round them, and work by way of them systematically.

2. Company and execution. Are you able to get stuff out of hypothetical and into actuality quick? The world is altering quick, the one option to succeed is to up your studying fee.

3. Product sense and style. You want a viewpoint on what ought to exist. If you do not have a viewpoint on the place this tech ought to go, you are in mission administration, not product administration.

What’s going to the PM job appear like in a yr?

Proper now most PMs spend time in the course of a distribution—between busy work on one finish and deeply human and inventive work on the opposite.

In a yr, I feel AI will deal with busy work. Whether or not it is Sprint or different automation platforms, each PM can be orchestrating a council of LLMs.

This frees up 80% of your time to speak to prospects and jam with good individuals to resolve issues.

Possibly we are able to get Peter AI and Morgan AI to align first and solely deliver within the people if needed.

Ha! That’ll be nice to see.

Thanks Morgan! In the event you loved this interview, comply with Morgan on LinkedIn and take a look at Dropbox Sprint.