Mike Knoop

mikeknoop@gmail.com (Mike Knoop) — Sat, 08 Feb 2025 00:00:00 -0000

Co-founder, CEO Ndea – an intelligence science lab
Co-founder, Board ARC Prize Foundation – a north star for AGI
Co-founder, Board Zapier – automation for everyone
Invested: Braintrust, Limitless, Julius, Cognition, Cartesia, After Thought, Climatix, New Theory, CTGT, Inception Labs, Sepal, Agemo
Grew up in St. Louis, now SF Bay Area
Email me: mikeknoop@gmail.com
Twitter: @mikeknoop

Interviews

2025: Introducing Ndea (gradient dissent)
2024: Alma mater commencement speech (university of missouri)
2024: Why AGI needs new ideas, ARC Prize launch reactions (sequoia podcast)
2024: Announcing ARC Prize (dwarkesh podcast)
2024: Why open source AI progress matters (no priors podcast)
2023: Zapier's AI-powered future (sacra)
2023: Category creation, APIs, & AI architecture (this week in startups)
2023: How Zapier employees use AI (greg kamradt)
2022: Founder's guide to building a $5B company w/ little funding (traction conf)
2019: Running remote-first product and design teams (yc podcast)

About

mikeknoop@gmail.com (Mike Knoop) — Mon, 26 Feb 2024 00:00:00 -0000

Co-Founder Mindset

mikeknoop@gmail.com (Mike Knoop) — Sat, 25 Nov 2023 00:00:00 -0000

Is it possible to do Great Work as a non CEO/CTO co-founder?

The temptation to be someone else is greatest for the young. – Paul Graham

Before starting Zapier, I read PG's other essay Keep Your Identity Small and now think there is an important connection: to do great work as a co-founder you must keep your identity small.

Writing about identity is tough without personal example, bear with me. I have fired myself from important roles 5+ times at Zapier. As engineer, platform manager, product & design manager, Chief Product Officer, and even as President of the company. I have changed roles so much it's hard to exactly define when one role ends and other begins. Each role could have been it's own career and I personally found it tempting to take off-ramps early on.

In contrast to strong identity of the CEO/CTO role, co-founders have to define it for themselves. To do great work as a co-founder, define your identity as “co-founder” with the job description “do the thing no one else can do.” [1]

When your startup is smaller, this often means figuring out execution holes in your startup and being hire #1 for lots of new jobs – doing the work directly in order to understand and the unique problems for your startup in that area before going to find someone who can durable solve them.

When your startup is bigger, this often means identifying the most important gap across the business in product, tech, strategy, or leadership – and again doing the work directly (because if anyone else could fix it, it wouldn't be the most important gap by defintion) and then figuring out a long-term plan, usually hiring or promoting.

Startups without cofounders must hire their way out. Obviously it is possible, there are many successful solo founder companies. Though comes with execution pain and risk (hiring is slow, hiring someone bad/mediocre is even slower) all the while the underlying issue persists. And when you get it right, the hiring doesn't scale; you've only plugged one hole and are about to find fifty more.

The main way to do great work as a cofounder is by consistently and repeatedly identify the most important gap and then solve it. In order to repeat, you have to replace yourself. And in order to replace yourself, you can't have your identity wrapped up in the role you created. The non CEO/CTO cofounder role at scaled startups is not super common. I think partly because you have to be a generalist but moreso because ego and identity get in the way.

I dropped out of grad school to help Wade and Bryan start Zapier in 2011. I'd never had a full time job and it seemed obvious that my co-founders would take on CEO and CTO roles. It is easy to be flexible about role on a small founding team.

As the team grew I felt an increasing pressure to justify my purpose. Not just for my sake but for the team as well! The most common piece of feedback I received from 10+ years at Zapier has been “do a better job defining your role”!

The first time I told the team I wasn't going to write code anymore, an early engineers asked “if you're not going to code, what are you going to do?” – I barely had a cohesive answer. I knew I was filling a lot of holes but I thought any explanation was unfulfilling. Not until years later could I articulate (by past example) what it meant to “do the thing no one else could do.”

Later I even fired myself from not writing code, to go write code again. I over stayed my role as Chief Product Officer role partly beacuse my identity was attached to it. I did eventually fire myself and got back to building in order to help Zapier become a multi-product company launching Zapier Transfer, Tables, and Interfaces in following years.

And ultimately I gave up the President role to go all-in on AI in Aug 2022, three months before ChatGPT was released. This time, I'd like to think I finally figured out how to keep my identity small to make big changes quickly.

The temptation to be someone else is greatest for the young. They often feel like nobodies. But you never need to worry about that problem, because it's self-solving if you work on sufficiently ambitious projects. If you succeed at an ambitious project, you're not a nobody; you're the person who did it. So just do the work and your identity will take care of itself. – Paul Graham

[1] Most often this is in context to your company, but sometimes is in context to the world. The latter cases are the most important things to identify and do.

Chief Hole Filler

mikeknoop@gmail.com (Mike Knoop) — Sat, 25 Nov 2023 00:00:00 -0000

AI Releases, March 2023

mikeknoop@gmail.com (Mike Knoop) — Wed, 15 Mar 2023 00:00:00 -0000

Wow, this was one of the most thrilling weeks in tech. It feels like a whole year of AI progress happened all last week (not to mention a banking crisis + resolution).

GPT-4 released
LLaMa running on consumer hardware
Claude released
Multi-model nat.dev playground released
Google AI showcase

GPT-4

I've used a version of GPT-4 (as did many, if you had access to the New Bing).

Compared to GPT-3/3.5 series, the key improvements for 4 are:

Significantly improved ability to understand, and reason about very long prompts. GPT-3.5 series allowed ~4k tokens. GPT-4 supports 8k with up to 32k in testing. This means way less “prompt engineering” and “model steering” is required to get the LLM to do what you want. See additional notes from me on the limits of current gen LLM “reasoning” here.
Multi-modal (text AND image input, text ouput). This hasn't been released yet, but OpenAI demo'd fully and correctly explaining screenshots, funny images, and memes. This is going to have big implications for software automation because we're about to see LLMs operating in “pixel space” to translate screens into structured language.

Other notes:

The GPT-4 technical report contains no details on parameter count, architecture, training time, etc. for “competitive and AI safety reasons”. Disappointing that only ~400 folks in the world know how the currently best-performing model works. We can still speculate.
GPT-4 training started 2 years ago and ended in August 2022. A look at historical GPT release dates:
- 2 –> 3: 16 months
- 3 –> 3.5: 18 months (InstructGPT)
- 3.5 –> 4: 15 months
The GPT-4 paper in mostly focussed on alignment and results. OpenAI says they sat on the trained model for 6+ months to do alignment work. Reinforcement Learning with Human Feedback (RLHF) was super important – the base GPT4 was not much better than GPT3.5 series on reasoning and standardized tests until RLHF was performed. Suggests parameter/training data scaling alone is not resulting in huge new capability unlocks but there is a ton of “latent capability” the LLMs are capable of once provided with a clear objective function.
Bing secretly using GPT4 was part of the rollout strategy to test quiet launch vs loud launch. OpenAI found it didn't help much with alignment goals so I'm not expecting many secret launches going forward.
The new “system” tokens are structurally going to be very important for programmatic use cases and preventing prompt injection.
Code gen is one area 4 does dramatically better than 3.5 series. Early demos show 4 can generate much longer, correct, runnable code zero-shot compared to 3.5. If GPT-4 can correctly generate code, it can also generate “no code”.

LLaMa running on consumer hardware

Several weeks above, Facebook announced LLaMa, an LLM they reported showed comparable performance to GPT-3 series models but much smaller in size (65B parameters compared to OpenAI's GPT-3.5 with 165B parameters). The model was leaked on BitTorrent (which is still live on GitHub which indicates at least some level of acceptance by FB).

Last week, Greg Gerganov open sourced an inference implementation (using minimal C++) using Facebooks' LLaMa leaked model weights.

LLaMa actually had several different model sizes from 7B parameters up to 65B parameters and importantly the 7B parameter version with instruction tuning is demonstrating coherent results! This is truly shocking and opens the door to running strong ChatGPT-like models on local consumer hardware with acceptable speed by converting the weights to int4.

And that's exactly what the internet has been sprinting towards the last week:

M1/M2 Mac (4 to 40 tokens/sec speed)
Raspberry Pi (1 token/sec)
Pixel 6 phone (5 tokens/sec)

This has several big implications:

Dramatically cheaper to deploy LLMs for lots of of use cases (pushes inference to client vs server)
Private inference now possible (eliminates sending sensitive data via an API to third-party)
No content filters (many users want raw model access vs sanitized corporate versions)

Each of the above alone would warrant a big deal. But together is quite revolutionary:

Llama LLMs running on M1/M2 consumer hardware def surprised me. Thought it would take another 12+ months to achieve that.
— Mike Knoop (@mikeknoop) March 12, 2023

Claude and a multi-model future

To date, the only realistic LLM API provider has been OpenAI. That's starting to change as Claude and Google both announced new model APIs for developers (and Claude shipped!)

In 12 months I expect there will be at least 4 or 5 major LLM API providers offering choice of model (and reduced platform risk) for the first time.

Nat Friedman released https://nat.dev/ last week which is a multi-model playground to compare LLM performance quickly in the browser.

New Product Development Cheatsheet

mikeknoop@gmail.com (Mike Knoop) — Fri, 26 Aug 2022 00:00:00 -0000

This is intended to be a useful knowledge base for anyone building new things, particularly new products. Assume you are building a new software product from zero.

By definition, new product ideas don't have “product/market fit”. This is, in my opinion, the canonical definition of product/market fit. The critical things to focus on first are:

What is the shortest path to P/M fit?
What is the opportunity cost for this idea?

And (1) is much more important than (2) because often big new things look like toys at first.

The shortest path to P/M fit is not a straight line because new ideas have to get tested by reality (eg. showing an early prototype to a potential user) and usually new ideas have to go through hundreds of adjustments as they hit reality before they can achieve P/M fit. So the right thing to optimize for, in fact the only thing to optimize for ahead of achieving P/M fit, is to increase the number of learning moments per week. A team that learns 5 things per week is going to move 10X faster than a team that only learns one thing every 2 weeks.

Stages

Pre-Launch
- Things to obsess about
  - First questions to answer
    - Who are you initially building for? Definition should be as close to a “SQL Query” as possible. The definition should help teams orient where to find the people we intend to serve in the world. Also important is “initially”. This will grow/expand over time. But better to be hyper focussed up front.
    - What is the minimum scope that will make those initial users in (1) love the product/service? Better to have 1 user falling over in love vs 10 users expressing mild interest.
    - If (1) and (2) work, how could this new product/service get big? This is where some vision setting comes in and acts as an important “opportunity cost” filter for ideas.
  - The only thing you should be doing is talking to users or building
    - Building customer acquisition counts too, Zapier built SEO landing pages before we even had a public product
  - Increase the number of learnings moments per week
    - Launch early – Whether it's an alpha, beta, etc. Launch early. If you're not embarrassed by what you launch, you've waited too long.
  - Do things that don't scale
    - Put your personal phone number on the landing page
    - Manually onboard every single user, use the software for them
    - Have them pay you manually, reconcile revenue manually
    - Manually find initial users on forums, existing user base, etc.
- Measures to pay attention to
  - Actual external users
    - UUs (beat zero)
    - Usage per UU (beat zero)
    - Retention per UU (beat zero)
  - Qualitative feedback from users of problem importance
    - Evidence of high motivation / excitement for problem to be solved
    - Look for revealed preferences, do they proactively ask for access every day after hearing about the new product?
- Time in stage
  - 6-12 months
- Team size
  - Just the founding team (“co-founders”):
    - Product/Technology/Design/Builder
    - Market/Customer/GTM
  - No more than one team (~7 people)
Post-Launch, Pre-P/M Fit
- Things to obsess about
  - Get to P/M fit: 10% WoW on usage/revenue is the bar to hit
  - Find a Go To Market acquisition channel that works
    - Most products that get big have a single dominant channel they acquire new users from (examples: SEO, content marketing, channel partners, WoM referrals, PLG referrals, advertising)
  - Keep doing things that don't scale
    - If you have 10 users on day 1, 10% WoW means getting only 1 more user by day 7. Very doable and you can still do super manual things to find/onboard new users.
    - Support your own product. Avoid the temptation to put people or processes between users and builders at this stage. Super critical because rapid iteration is still necessary.
  - Willingness to pay
    - Do not give it away for free at launch. Charge something. Very important as it reveals true problem importance and value.
- Measures to pay attention to
  - Growth rate
    - Consistent 10% WoW on usage/revenue is the bar to hit for P/M fit
  - Usage
    - UUs
    - Usage per UU (going up as you add features/value)
  - Retention
    - Weekly Cohorts (increasing retention over time w/ new features?)
- Time in stage
  - 12-24 months (if going beyond 12 months, avoid the “one more feature” trap. Make significant pivots or de-invest.)
- Team size
  - Up to one team (~7 people)
Post-P/M Fit
- Things to obsess about
  - Growth. Easy to get distracted at this stage. Keep making sure the #1 priority is to increase growth.
- Measures to pay attention to
  - Growth rate
    - Hold 10+% WoW as long as possible. Silly thought experiment, if you do this for 36 months in a row your new product is bigger than Google.
  - Monthly metrics
    - MAU/WAU/MEA (some time series measure of engaged user)
    - ARPA (average revenue per account, increase it with tiering)
    - NRR (net revenue retention, goal is to get 100+% monthly NRR)
  - Pricing and packaging
    - With significant users and revenue, can start optimizing price points and models to figure out the elasticity curve.
- Time in stage
  - Until you run out of ideas and growth stalls.
- Team size
  - Throttle off. Grow the team as necessary to drive usage/revenue growth

Mike Knoop

Mike Knoop

Interviews

Archive

About

Co-Founder Mindset

Chief Hole Filler

AI Releases, March 2023

GPT-4

LLaMa running on consumer hardware

Claude and a multi-model future

New Product Development Cheatsheet