• Colleges and Upward Mobility

    Attending college is still a major force in upward mobility for children with parents in the bottom quintile of income to reach the top quintile of income. A recent working paper from NBER looked at the role of colleges in explaining upward mobility outcomes.

    Looking at the top 10 colleges for upward mobility in the US I was surprised to see my alma mater Pace University (which also happens to be one of the only for-profit schools on the list) and SUNY Stonybrook (roughly 20 minutes from where I grew up).

  • Context Ascribes Value

    Context matters in the way that people recognize value. For example, a world-class classical musician performing in the street is unlikely to be recognized for their brilliance compared to performing in Carnegie Hall. When they are on the street, they are a street performer barely noticed by rushing commuters. When they are in a concert hall, they are in the context of classical performers compared to every other classical musician in the world.

    From Obviously Awesome (literature notes).

  • Obviously Awesome (Literature Notes)

    How to do positioning

    1. Make a short list of your best customers
    2. Form a positioning team Must be driven by the leader of the business and the leaders of each business function or it won’t be adopted
    3. Let go of baggage
      • Everyone must be on the same page about what positioning means, each component of positioning, and styles of positioning depending on the market
      • Get agreement that the product was created with a certain market and audience in mind but may no longer be best positioned that way
    4. List true competitive alternatives from the customers' point of view
      • What would your best customers replace you with if you didn’t exist?
      • Aim for 2-5 groups of alternatives
    5. Isolate unique attributes/features that make you different and better than alternatives
      • Focus on provable facts rather than unqualified values like “outstanding customer service” or “ease of use”
      • Concentrate on attributes that customers care about when evaluating whether or not to make a purchase.
    6. From attributes, create value themes
      • Feature -> Benefit -> Value
        • Feature: something your product does or has
        • Benefit: what a feature enables a customer to do
        • Value: how the feature maps to the customers' goal
      • Group into themes from the perspective of a customer
    7. Determine who cares a lot
      • Segment the features and values by best-fit customers by asking who cares and why?
      • A segment needs to be big enough to meet business goals and has important, specific, unmet needs that are common to the segment
    8. Find a market frame of reference that puts your strengths at the center and determine how to position in it
      • A market needs to be something that already exists in the minds of customers and highlights your strengths
      • Abductive: choose a market category by isolating key features/values and asking, “What types of products typically have those features?” and “What category of products typically deliver that value?”
      • Adjacent and growing: look at a market adjacent to where you have been positioning yourself—ones where there are already blurry lines and feature overlap
      • Ask customers: be careful with this, they may have already tried to make sense of you and failed, they will only try to position you in markets linked to their industry and job function
      • Positioning styles
        • Head to head: aim to be the leader in a category that already exists. If there is a leader, beat the leader and convince customers your solution is best. Even better if there is no leader.
          • You don’t have to teach buyers and can rely on what they already know
          • You need to fit within their existing definition if you want to win
        • Big fish, small pond: Win a well-defined segment of the market, serving an underserved segment of the market whose requirements are not met by the current market leader, then expand until you can compete with the established leader
          • It’s much easier to gain traction, be more targeted, and be accretive (each subsequent customer is directly applicable to the next, word of mouth)
          • Subsegment needs to be easily identifiable, it should be easy to make a list of prospective buyers
          • It has to be possible to demonstrate that there is a very specific and important unmet need
          • You can fully solve their specific pain better than the leader
        • Create a new game: create a new market category—prove the new category deserves to exist, define the parameters of the market in customers' minds, position yourself as the leader in it
          • Should only be used when you have evaluated every possible existing market category and conclude you can’t position there
          • Can only work if you are large and powerful enough to get the attention of customers, media, and analysts to make the case for why a new market category deserves to exist
          • New markets emerge from change—new technology, economic changes, politics, preferences, or a combination of those
          • Your product must be inarguably new and different from what exists
          • It often takes a long time to do well, but if successful it can be huge
    9. Layer on a trend (maybe)
      • If the trend doesn’t reinforce your positioning, it can muddy the waters (cool but confusing, lies, good but boring)
      • The trend needs to clearly connect to your product
    10. Capture your positioning and share it across the organization

      • Write a document with enough detail that it can be used by marketing, sales, and product (1-pager, positioning canvas)

      • Positioning Canvas

    11. Create a sales story
      • Define the problem your solution was designed to solve
      • Describe how customers are attempting to solve the problem today and where current solutions fall short
      • In a perfect world, the features of a perfect solution would look like…
      • Introduce the product and position it in the market
      • Talk about each value theme in more depth about how the solution enables value
      • Wrap up by addressing common objections, case study, current customers and next steps
    12. Make a messaging document Write out the messaging that every new campaign or material should be reviewed against
    13. Product roadmap and pricing Adjust the product feature set to align with the market category and get your pricing inline with market how the market expects it to be priced
    14. Check in on positioning every six months or after a major event that could change the competitive landscape (e.g. a new large competitor, regulatory change, attitudes, technology)

  • Make Two Offers

    Making two offers with different tradeoffs sets the framing of the negotiation and reduces the amount of back and forth. For example, making two job offers to the same candidate—one high salary but lower equity and one lower salary but higher equity—and asking them to pick—eliminates a lot of the need to negotiate (which most people don’t like to do). Using this strategy makes it more transparent to the other party how to model the deal and empowers them to decide.

    See also:

    • If negotiation is a coordination problem then making multiple offers at the same time accelerates finding the Schelling point
    • The poverty of compromise
    • Salary negotiations are more balanced between parties (you’ve probably done several and can ask others), you should be skeptical of this strategy in a negotiation with high information asymmetry

  • Do the Hard Thing Perfectly That No One Wants to Do

    A tried and true way to build a large business is to do something that is difficult really well that no one wants to do.

    There’s actually a lot of work people want to do. Things that feel rewarding, creative, and fulfilling.

    Then there is work that no one wants to do. Doing your taxes, managing a cap table, paying bills, staying compliant, etc.

    An advantage that businesses which do what no one wants to do themselves and has to be done perfectly (stakes are high e.g. fines, jail) is that there is built-in marketing and less competition. Customers are reminded they want someone else to do it every time it comes up—when they get a letter in the mail from a state agency, when they get a penalty, when their lawyer gives you homework, etc.—enforcement is marketing. Less competition gives room to make more progress in the market before it becomes an attractive business—if you survive long enough, when the market emerges you were there the whole time.

    See also:

    • Combined with hiring people that care a lot and there is a good shot of making something significant (some would argue Stripe’s innovation is getting smart people to work on schlep-y problems)

  • Zapier Natural Language Actions API

    The Zapier NLA API solves a major problem for large language models—the ability to interact with real systems. Rather than a developer integrate with every possible service, they can integrate once with Zapier and run every “Action” the user has authorized using natural language instructions.

    One thing I noticed after testing it out is that the underlying action materially impacts the quality of results. For example, the Slack API is fairly straightforward to parse the instructions into the correct action (e.g. post this message to this channel). Whereas something more complicated (with an API design that is not great or very abstract) like the HubSpot API fails on the simplest instructions (e.g. “find XYZ” to the Find Company action).

    The primary issue I see here is that a lot of control is deferred to the NLA instructions which the developer has no control over. This is going to lead to really odd hacks where you need to wrap the instructions to the NLA API with another language model to augment the instructions to something more explicit creating a bad interop problem (calling APIs directly would be better at that point).

  • Venture Predation Is Predatory Pricing for Startups

    Predatory pricing claims are largely ignored by courts, but a version of it continues to happen with venture backed startups. A recent paper that is making waves in tech circles called Venture Predation shows how businesses like Uber, Bird, Moviepass, and more use large venture capital investments to undercut competition, build monopolies, and harm consumers.

    The interesting part about this flavor of predation is that the business does not need to actually work to be successful for investors and management—merely the promise that it could someday work is enough. For example, “We’ll subsidize ride-sharing to capture market share quickly and then build self-driving cars later to recoup the losses and raise prices later.”

    Even more interesting is what the paper says about the key innovations of venture predation to get away with predatory pricing:

    Venture predation adds three innovations to traditional predatory pricing: (1) motivated financiers motivated to fund predation, (2) the secrecy of private companies, and (3) the opportunity to cash out before recoupment.

    I tend to agree with the authors that the confluence of factors around little/no enforcement against predatory pricing and generally how startups work gives rise to this playbook. However, there are plenty of other examples from large established businesses (probably monopolies in their own right) like Amazon and Walmart that can use their pricing power and economies of scale to simultaneously crush competition and fund growth with short-ish-term losses.

    See also:

  • Large Durable Business

    People often ask me in interviews where I see our company going in the coming years (which is more of a meta question asking about the exit strategy). I always say the same thing: the plan is to build a large durable business because if you do that, every option is available to you.

    Building a large durable business is exceedingly difficult. To be “large” you need go after a big enough market or one that’s growing quickly. To be durable you need to build something that is defensible (a moat or a lasting competitive advantage) and diversified (resistant to headwinds). Finally, there needs to be a business in there (which is surprisingly overlooked) that generates revenue for less than the cost it takes to deliver value.

    If you happen to build one of those rare, large durable businesses, you can sustain the business without investors (the small version of this is “ramen profitable”), pay dividends, raise more capital, IPO, or sell it. Not all of these options are available if you fail to build a large durable business.

    (Note that a “sustainable business” is not the benchmark. There are many kinds of sustainable businesses that could run forever, but not necessarily large ones which is rather the point of startups.)

    I borrowed some of this from my time working at Stripe where this concept was talked about somewhat regularly.

  • How to Decide if AI Tools Can Be Used at Work

    Advancements in AI powered tools can greatly improve productivity but many companies have taken steps to limit or outright ban the use of OpenAI’s ChatGPT, GitHub Copilot, and others. What are they concerned about and how should you decide if it can be used by your company?

    Risks of AI tools in the workplace

    Because large language models are very big and resource intensive (this is changing), they need be run on servers rather than on device. Since these models work on text, that means transmitting a lot of potentially sensitive information over the network. To my knowledge, none of the major AI platforms are offering end-to-end encryption.

    There are also privacy and IP concerns. If information sent to be processes is mishandled it could leak important IP or trade secrets. Apple recently banned ChatGPT and I suspect that is the reason. I’m guessing there are also legal concerns about ownership if AI generated output ends up in a company’s IP.

    How to decide

    The value of AI tools in the workplace is productivity. If GitHub Copilot can improve developer productivity even a small amount, it would be well worth the return given the cost of engineering. On the other hand, there are real risks.

    These risks can be managed by thoughtful policies, training, and controls.

    1. Do not allow sensitive customer data to be sent to AI tools over the network Mitigations might include deciding which teams can use e.g. ChatGPT, creating training on how to use AI tools, build an in-house wrapper around LLMs that can detect certain data like credit card numbers and IDs
    2. Avoid using coding tools that require access to the entire codebase Mitigations might include only allowing local language models, ensure all secrets and API keys are encrypted or not store in version control, ban Copilot but allow engineers to use ChatGPT for code help
    3. Buy the enterprise version of these tools, ban personal use tools Many providers realize that stronger guarantees about data use and storage are necessary for businesses. Coupled with banning any personal AI tool usage, it could provide more privacy and security.

    This was just a quick list of ideas but it seems like more nuanced approaches can be taken to balance the risk and reward of using these tools. One big thing is missing though—what new failure modes and risks now exist as a result of using these tools? I guess we’ll find out soon enough.

  • Build a Cross Platform FAISS Index

    Building a FAISS similarity search index is not cross-platform. If you save the index locally using aarch64 and try to load it an x86_64 environment it will not work (for me, it loads an empty index). Using docker, you can save an index built locally so it’s compatible in other architectures.

    To do that, run a container of the image which has your setup using the --platform flag (matching the architecture of the servers you want the index to be loaded) and run your code to build and save the index (using the FAISS save_local method). The index files generated in the container will get synced to your local machine via the volume mount so you can use them how you need (e.g. check them into version control, store them in S3).

    docker run --rm -it \
       --platform linux/x86_64 \
       -v $PWD:/my/container/path myimage:latest \
       sh

  • Productivity Is Bounded by Decision Making

    At a certain point, optimizing productivity becomes optimizing for speed of decision making. After all the tools, shortcuts, and hacks, that build up raw speed to get tasks done, you’re left with the cognitive load of decision making. That email you received? It’s a decision disguised as a reply. That Slack message that remains unread? You’re procrastinating because a decision needs to be made that you don’t want to confront.

    So many productivity frameworks and tools rely on cleverly hiding tasks in queues so that you can do them later. Try being more decisive and see what kind of impact that has on your to-do list.

    See also:

  • Clarity Is One Number

    Making complicated things seem simple involves abstracting over reality in such a way that is clear and actionable. Often times, that means reducing things down to one number going up or down. People are drawn to (fixated even) clarity of a single number going up or down.

    For example, your weight captures a high degree of nuance at low fidelity—it could go up or down for a myriad reasons—but provides clarity in a way that tracking dozens of bio-metrics does not. If it starts to go up, you might look at it with concern, if it goes down, you might celebrate this as a victory.

    We see this desire for one number everywhere. A stock price that grossly encapsulates a company’s value and the market’s psychology. The score in a baseball game indicates who is winning and who is losing. The Earth’s average temperature rising indicating catastrophic climate change.

    See also:

  • Bobby Bonilla Deal

    The New York Mets made one of the worst deals in sports history. From 2011 to 2035, the Mets have and will pay Bobby Bonilla, a baseball player who has long since retired, $1.19MM every year.

    A “Bobby Bonilla deal” is one that results in substantial compensation being paid with no value being generated at all for a long time. You can tell it’s a Bobby Bonilla deal if you ask the question, “What value will the other party bring 10 years from now?” and the answer is, “Nothing but they will still get paid handsomely.”

    This might sound a lot like royalties, but I assure you it is not! With royalties, there is a tangible, renewable, asset that has value and can be traded. Its value could decline but so would usage and therefore royalty payments. On the other hand, a professional baseball player is a depreciating asset that, by definition, goes to zero.

  • Bluesky Will Never Be the Cozy Web

    Bluesky is having a moment where all the users are new, the content is shitpost-y, and it all feels very lively. They’re also getting their first taste of unsavory people joining, ruining the collective bubble.

    The cozy web is incompatible with large-scale social networks because the cozy web needs to be small and social networks need to be large. As a result, there are no controls that could be put in place to simultaneously build a large public social network and make it free from bad people.

    It’s unfortunate because Bluesky is (currently) really fun and whimsical, but I can’t help feeling we’re watching another social network speed run through everything learned about social media and content moderation over the last decade.