James, Stephen and I started working on Manifold in late November. Now that we’ve passed the halfway mark of 2022, how has Manifold performed overall?
Things are loosely categorized by:
- Product (the actual thing we’ve built)
- People (the team, users, and community around Manifold)
- Process (the governance and day-to-day execution; how we make decisions)
What we did well
- Product: fairly well-designed product experience, tapping into a lot of high-level interest in prediction markets
- Things our users say that we’re doing well:
- It’s really easy to get started on Manifold; it’s free, and very permissionless
- A lot of interesting/novel use cases on the site
- The design of the app feels very modern
- We ship things really quickly
- Subjectively, I would rate Manifold a 7/10 on overall product polish and usability, which puts it somewhat below “top-tier Silicon Vally consumer-tech startup”, but far above “average crypto website or EA tech project”.
- Random grades for comparison: Metaculus is 5/10; Substack is 6/10; Facebook is 7/10; Kalshi is 7.5/10; LessWrong is 8.5/10, Notion is 9.5/10
- Writeups have been fairly positive so far; some of the people I most respect in the world had nice things to say about us
- I’ve done like 5-10 side projects/startups but Manifold is the first time someone sent me a physical thank-you note
- Literal code contributions, both on API and directly to the site
- People: our core users are highly engaged; and we’ve gotten interest
- Real money coming in via user payments
- People putting their money where their mouth is
- One of the things I got wrong the most was “why would people pay for internet points?”
- A community is when people talk to each other, not just to you
- One of our earliest markets: “Will our Discord be a full community by February 1st?” - blew way past my personal expectations
- Users chat on Discord, and crucially on Manifold itself. In depth discussions on a wide variety of topics, with people dropping in expertise
- A lot of partnerships
- Private instances for the Atlas Fellowships to let students get familiar with the mechanics of prediction markets
- Markets being used for scientific research: https://research.bet/, Spencer Greenberg on psychology study replication
- Copying over information from Metaculus
- Onboarding different creators in the rationality space: Zvi, Scott, Aella, Sarah Constantin
- Process: we’re iterating rapidly, and trying lots of different things
- We’re conducting a lot of what feels like fundamental research and innovation
- Free Response Markets, Numeric Markets v1
- Building an automated market maker (traditionally a “crypto” idea) on Web2
- Prediction Markets for Charity
- Intersection of economics, gaming, and social media
- Eg no other prediction market integrates discussions the way we do
- Being willing to launch quickly
- From the beginning: the Manifold MVP was up in 3 weeks.
- More recently: Groups, Notifications shipped from scratch in the span of a couple of weeks
- Permissionless innovation is core to the Manifold site itself (make any market you want!), but also to feature development within Manifold
- Instead of trying to persuade us with words, just show us with actions
- Internally: even if a founder is skeptical, you can just go build the thing
- Prefer showing a demo to writing lots of design docs
- Mantras:
- “Don’t let perfection be the enemy of good enough”
- “Easier to ask forgiveness than permission”
- “If you’re not embarrassed by your v0 you took too long to launch”
- Make iteration afterwards very easy by aggressively making the core developer experience good. The time from “oh we should fix this” to “hey this is now fixed and shipped” should be on the order of minutes
- Part of this is killing things that aren’t working; distracting from the core userbase
- Process: Being radically open
- Our codebase, metrics, design docs, salaries are all public
- You can chat with any of us on discord
- You can book a time to chat with me anytime
- Benefits of this approach:
- Allows people who really like what we’re working on to engage with us
- High trust in our decisionmaking; we have nothing to hide
- Reduces friction in internal communications, less silo-ing, fewer meetings to sync up
- Forces us to act with integrity and structure scaleable processes
- Drawbacks of this approach:
- Can be very distracting (on the margin, Discord is a time sink, maybe Manifold itself too)
- Imposing legibility might reduce throughput? Sense that we’re spending a decent amount of time eg reviewing code, getting consensus
- Process: Operating as a remote startup, with frequent in-person onsites
- Austin is in SF; James and Stephen in Austin, TX. And we’re too stubborn to want to move to a different place.
- So: Manifold has been remote since inception. And our processes are set up to make this streamlined. Some principles:
- Async is better than synchronous. Notion is better than Discord, is better than having a meeting
- Weekly 1:1s to stay in touch between most pairs of people; provides a space for concrete feedback that wouldn’t be otherwise delivered
- There’s still no substitute for focused getting-to-know-each-other time in a real physical location. So every 1-2 months the team gets together for a week or two, in different locations. So far, we’ve done Austin, SF, Bahamas, Mexico City; planning on SF (again) and DC.
What we could do better
- Product: markets aren’t being used to make important or core decisions
- Strict Futarchy: very few markets being used to decide outcomes of things
- A lot of possible experiments in this space (eg creating a futarchic agent to play games) that we’re not running at the moment
- Could design some things to make this use case a better UX, or more capital efficient (aka spend less M$ subsidizing markets to get good results)
- Loose Futarchy: the predictions from our markets aren’t yet trusted as an input into important decisions
- Not even for us! Most decisions about Manifold, and especially important ones like “who should we hire”, are not primarily conducted via prediction market
- This is a pretty big failing, IMO
- Nobody says: “this Manifold market really changed my life”
- Long term markets are really hard to incentivize properly
- Like with most posts/articles, there’s typically a burst of interest when the question is posed, and then a dropoff of interest over time
- Tried loans; some execution problems with our first iteration. (Will likely revisit). But also a lot of other ideas eg margin lending, separate long-term currency.
- Product: Social is a double-edged sword
- People: Not a lot of growth in our usage numbers
- Process: task prioritization and team coordination is hard, especially with so many possible use cases for prediction markets
Where we got lucky
- Process: Fundraising was somewhat easier than expected