🎬 And that's a wrap!
Some snapshots from our full house one-night speakeasy with @KoyalAI
From 1,000+ RSVPs, we hand picked 160 YC Startup School attendees and invited them to a secret location. They stepped into an evening of cinematic production, drinks matched to their Github activity, and our first limited-edition perfume. Be on the look out for our next one ;)
From day one, our mission has been to build the infrastructure needed for an agent-native world.
At NY Tech Week, we gathered experts from @OpenAI, @google, @stripe, and @paysponge to share their thoughts on building towards a true A2A economy.
We discussed:
◆ Does the word agent even mean anything?
◆ Who will win the agent communication standards wars?
◆ What's left to build before agents can become "sovereign"?
Watch the full talk ↓
Last week at NY Tech Week, founders and builders filled the room for our BYOA (Bring Your Own Agents) workshop.
Everyone brought new ideas and their own agents to showcase and try new tools from us and our partners.
Big thank you to our co-host @Mercury, and to @paysponge and @subsysdev for joining us!
NY Tech Week is a wrap!
Dedalus closed out the week with Run(way)time, a fashion show celebrating the launch of our merch line and the future we are building with Dedalus Machines.
We brought together founders, builders, artists, designers, and investors to explore creative use cases for AI agents while celebrating a staple of dev communities: merch, reimagined through an elevated runway-ready lens.
Guests received personal agents through @AgentPhoneHQ to play a live game, watched our new merch collection on the runway surrounded by visuals from @fuserstudio and other creative technologists, and enjoyed art and music throughout the night.
Thank you to everyone for the overwhelming interest and support. We’re grateful to everyone who helped make this event possible.
why are we still babysitting agents in 2026?
we’re hosting BYOA(agent) during NYTW
with @mercury@paysponge@subsysdev
bring your agent,
we’ll help you level it up with tools + infra
limited spots left. merch included ;)
over the past year, we’ve created a thriving community of builders. naturally, many became great friends.
our hoodies have been spotted in sf, nyc, pittsburgh, boston, princeton, yosemite, hawaii, vegas, salt lake city, london, berlin, paris, bombay, shanghai, tokyo, seoul, and more.
> if you have the hoodie, it's time to scan the barcode :)
Dedalus Labs turned 1!
To celebrate, we’ve hidden a few easter eggs on our website. Look in the right place and you might find a gift.
Hint: it’s time to 𝚎𝚡𝚎𝚌 your biggest wishes
< Dedalus Archives > drops next week.
We're gifting 10 users our new collection.
To enter the raffle:
◆ Create a Dedalus account
◆ Join the Dedalus Machines waitlist
To claim extra entries:
◆ Like + Repost this post
◆ Reply with the piece you want most
Winners will be announced on May 1st.
At HackPrinceton last weekend, it was great to see builders ship amazing projects on Dedalus. We also hosted a session discussing Dedalus Machines and why persistent, scalable infrastructure matters for agents with a packed room.
Thank you to everyone who built with us, came by our talk, and grabbed our swag!
I've been recently doing the following:
- Prototyping w/ Opus 4.6
- Code reviewing w/ GPT 5.4 xtra-high-fast
Opus 4.6 invariably writes a buggy first draft, but I've found that iteration speed is super important.
GPT 5.4 is the "heads down" staff engineer who produces exceptional work when given uninterrupted budget to focus. It will come back two hours later with a polished solution, but at the cost of iteration speed. Sometimes, it'll miss the mark, which effectively kills all momentum.
Opus is the energetic junior engineer: young and full of enthusiasm ... but forgets to read the docs sometimes. Debugging its initial attempts takes up another several hours before I resign back to GPT 5.4.
For the MOST difficult problems we've been working on at @dedaluslabs, I've found that it's better to be directionally correct (~80%) for the first draft, then fix the remaining bugs (~20%) incrementally along the way.
My longest days were when I exclusively only used GPT 5.4 xtra-high with little to no feedback from the model.
Yet another reason for model choice and a vendor agnostic future.