GTM Lead @ Fundamental | Ex-JPM | https://xitter.cf/t.co/BzYpc970Cd
Joined August 2021
- Tweets504
- Following281
- Followers176
- Likes943
Pinned Tweet
Sharing Knute - an AI agent that manages your fantasy football team.
Was a fun side project and learned a ton building it.
Will share more about how it all came together. Would love if you tried it out! All feedback is appreciated!
knute.ai
My dot just became the best onboarding tool of all time.
I gave my dot instructions to answer any questions our new hire (ex-jpm IB) has and allowed him to send emails to my dot.
My dot then responded to him sent him html files and gave him run downs on EVERYTHING he wanted to know
I’d like to file a p0 complaint with google. The 25x mfa on everything is the only bottleneck in my life that triggers every cuss word I know
Ok not true but seriously mfa on devices and accounts I use daily is obnoxious and it blocks my friends (agents) from being helpful
nice benchmaxxing by meta - can’t wait to see @nicochristie xbench results in a week
will be immediately moving mobile service when this releases if it’s fast
We announced an agreement to acquire a nationwide low-band spectrum license portfolio that will pave the way for @Starlink to become a major mobile carrier in the US.
With this new spectrum and our Gen2 constellation, Starlink Mobile can ensure Americans have access to high-speed mobile broadband no matter where they are → spacex.com/updates#starlink-…
🏎️🏎️🏎️
Astra Ultrafast is here in Shortcut.
Faster financial modeling, with one click.
Watch a quarterly earnings model update finish 3.5× faster in this demo: 1 minute 15 seconds vs. 4 minutes 24 seconds.
Select Astra → Ultrafast below the chat box and try it on your next workflow at shortcut.ai.
David Maginn retweeted
$NKE earnings are up next. Using Shortcut, I’m predicting $0.45 normalized EPS vs. the Street’s $0.44.
After adjusting my forecasting approach to account for company size and historical earnings beats, I've been closer than the Street on 7 of the last 8 results. $MKC was the only miss in that stretch.
There are of course limitations to these results:
1. It’s a small sample, and all 13 companies that have reported beat consensus estimates.
2. This cohort beat consensus ~80% of the time over its previous six quarters. That makes predicting a beat easier, but getting closer than the Street still requires estimating a company's size since overshooting can lose.
These are encouraging results, but there's more tests ahead with a fresh batch of companies next week. $NKE is our final result this week.
Predictions, results, and the financial models are all public, check them out here:
beatthestreet.live.
I spent today setting up Dot and I think this is the exact tech I’ve been waiting for.
If you’ve already built workflows, skills, and projects in Codex, you can effectively teach Dot how to use them. It needs guidance at first, but once you show it how you work, it starts to click.
A few things have made the biggest difference:
Projects matter. Work should happen inside the right Codex project, where the files, instructions, skills, and prior work already live.
Reusable workflows > repeated prompts. If I’ve already built a process or skill for something, Dot should use it instead of reinventing the task every time.
Dot should coordinate the agents. I give it the outcome. It delegates, checks the work, and brings me back something finished.
History is part of the instruction set. Before asking me a question, it should search past conversations, emails, docs, and code for precedent.
“Done” should actually mean done. Not just a draft. Attachments included, checked, saved in the right place, and ready for the next step.
Access ≠ permission. Give it enough context to work independently while keeping approval gates around things like sending messages or signing documents.
Feedback should improve the system, not just the output. Every correction should teach it how I want work approached going forward.
The other massive unlock: I can message or call Dot from my phone, talk through an assignment wherever I am, have it coordinate work across the projects on my computer, and come back to something ready for review.
This is starting to feel much closer to an actual operating layer for my work than another chatbot.
David Maginn retweeted
An earnings eval is meaningless if an agent can just Google it or spout it out from memory.
That was the challenge I ran into building Streetbench. I first tested AI earnings forecasts on roughly 100 companies in a previous post. For Q2 2026, I expanded to 200 companies, set a pre-report cutoff for each one, and used Exa Snapshot for the agents’ research. I also audited the agent's research trajectories for leaked results.
I tested 9 combinations of model and agent combination: Shortcut, Codex, Codex+GPT Researcher as harnesses along with 3 models Luna 6, Sol 6, and Astra all high effort. Shortcut with Astra had the lowest average EPS error, $0.1203 per share versus $0.1467 (18% reduction) for the final Street consensus.
What's most exciting is the progression of scores within Shortcut. Error fell from $0.1316 with Luna to $0.1274 with Sol to $0.1203 with Astra. Better models produced better estimates in this setup, which is an encouraging sign for what these agents could do as capabilities improve
I've published the 200 cases, cutoff times, methods, results, and scorer on GitHub. Try another model or agent architecture and see if you can beat Shortcut with Astra.
github.com/fundamental-resea…
I'm also posting forecasts before earnings at:
beatthestreet.live/
Claude can now help you build evaluations and hillclimb on them.
In this article, we share guidance on eval design & skills that Claude Code can use to improve your applications.
claude.dev/blog/automating-e…
David Maginn retweeted
if i ran a major league baseball franchise i’d feed every stat imaginable into frontier ai models & basically delegate the roster construction + entire game management layer to them. no traditional manager & no gm making baseball decisions by gut ever. the manager & gm are both ai agents.
for every game every pitch would update the model including exit velocity, pitch velocity, spin, fatigue, matchups, positioning, bullpen availability / usage, etc. & it would continuously simulate thousands of scenarios & make the highest expected value decision in real time.
then over a 162 game season it watches every game afterward through precision cameras, grades every decision, learns from every outcome, & gets better.
baseball might be the perfect sport for this. it has discrete events, absurd amounts of data, repeated situations, & a gigantic sample size.
billy beane found mispriced players but this finds mispriced decisions too on every layer.
can some billionaire please buy an mlb franchise & let me run this experiment pretty please?
this would be the houston astros on steroids. imagine ai winning a ring?
cc @JoshuaKushner
David Maginn retweeted
Shortcut cut median EPS forecast errors by 20% in my backtest. Now I’m publishing this week's 14 forecasts before earnings to test it head to head against Wall Street.
One fair point of feedback on my last post was that a backtest is difficult to trust and allows for overfitting. Even with controls for information leakage, forecasting before the results exist is the real test.
I’m starting a weekly report of my earnings forecasts using Shortcut, beginning with companies reporting this week, including Micron, Nike, CarMax, and Accenture.
For each company, you’ll be able to see my EPS estimate alongside the Street’s and inspect the financial model and reasoning behind it. After earnings, I’ll add the actual result and compare how close each forecast was. The original forecasts, wins, and misses will stay on the record.
I'm treating this project as a stepping stone into a world where investors use agents as a starting point for all research and modeling. Publishing the models also lets us look beyond the final EPS number and see which assumptions the models made hold up.
Three of my biggest differences from the Street this week:
• CarMax: $0.62 vs. $0.72 Street (−14.2%)
• Nike: $0.40 vs. $0.44 Street (−8.8%)
• Progress Software: $1.65 vs. $1.52 Street (+8.6%)
Rounded EPS estimates, percentage differences calculated before rounding
Explore all 14 predictions and the financial models behind them here:
beatthestreet.live/
One of my favorite days in my 31 years
Had no PTO left, first year out of college.
Took a flight to JFK got in at 6 PM after 1.5 hour delay.
Uber line was insane - hundreds of people waiting, no cell service.
Knocked on a drivers window and said I’ll give you $100 cash if you get me to Yankee stadium by 7 PM
Walked in the stadium during the Anthem, found my Dad, and Hicks hit this home run ~10 minutes later.
I couldn’t talk for a week after this one.
Opus 5.5 literally just runs for days without interruption. It’s on like a 3 day task and won’t stop… 500+ agents in.
chances the president just found Claude code and this was his reaction are non 0
Many people think that the words “Artificial Intelligence” are inaccurate, and very ineloquent, relative to AI, or Artificial Intelligence. A far more elegant and accurate description of this new phenomena would be Superior Intelligence (SI) or, Extreme Intelligence (EI) or, Supreme Intelligence (SI). This is a Poll, and I would appreciate everybody voting! Which is the best name for this ever growing “Revolution?” President DONALD J. TRUMP
