
vibemaxxing @orchid_hq pho @PhotonHQ
Photon HQ
Joined March 2022
- Tweets1.7K
- Following1.2K
- Followers617
- Likes4.5K
Darshan 🌸 retweeted
We are organizing the world’s first Phone Use hackathon.
10 hours with 3000$ in prizes.
The mobile era has come.
So we set only one rule: everything has to be accessible from a phone.
We are backed by Phone Use from @abdullababakre, @Replit, @LangChain, and @ElevenLabs.
> Friday November 13th.
If you’re not part of the 0.1% i think that you shouldn’t apply to the link in the first comments.
Darshan 🌸 retweeted
Payroll software from 2000 has no integration and never will. On Maritime, your agent gets its own computer with that software installed, and runs payroll the way a person does: open the register, fix the exceptions, file the period.
RT @DavidPawlan: Super pumped to announce 6 AI assistant legends who will be given the spotlight at the upcoming AI Assistant SF meetup on…
Darshan 🌸 retweeted
we've officially launched Maritime on YC!
when we started Maritime in march, giving every agent its own VM was considered wasteful and stupid
then grokbot, muse, and instinct came out.
ai agents should be fully autonomous, and we made this a possibility through our platform
go deploy your agents on @maritime_sh, go get those users
Darshan 🌸 retweeted
🇬🇧 UK numbers are now live on AgentPhone!
You can now give your AI agent its own UK phone number.
To celebrate, we're giving away $10 in free credits.
Just sign up at agentphone.ai and DM us your signup email to claim yours!
Darshan 🌸 retweeted
bruh i really thought it was a real person...
give it a year and your support calls, job interviews and tutors all have a face like this, and the first thing anyone asks on a video call is "are you real?"
Introducing Griffin, the first model to pass the video Turing test.
48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video.
It’s the first Human Interaction Model (HIM).
Readers added context they thought people might want to know
The 48% figure and "video Turing test" claim are from Tavus's own study of 54 one-minute calls, not independently verified or using a standard protocol. Griffin-Lite leads NVIDIA's VideoFDB benchmark on their public leaderboard.
cellcog.ai/blog/tavus-gri…
research.nvidia.com/labs/amri/proj…
tech-ish.com/2026/10/02/tav…
Darshan 🌸 retweeted
The @ycombinator virtual office simulator is really good.
@garrytan you forked yourself!
Darshan 🌸 retweeted
text your friends and family stocks on @base
Darshan 🌸 retweeted
I wanted to build an AI agent that does something quite different
Not another chatbot, not another “AI assistant” that gives you an answer in 5 seconds and definitely not an agent that simply agrees with whatever you tell it
.......so I built Conclave - a multi agent decision system that lives inside iMessage, built on @PhotonHQ
Here's the idea - You bring the problem, Conclave brings the room
When you have a difficult decision, you usually ask one AI, “What should I do?”
The problem is that one perspective can be wrong.
It can miss an assumption,
It can overlook a risk,
It can confidently recommend something based on incomplete information.
but Conclave takes a different approach.
Instead of asking one AI to make the decision, it assembles a panel of specialized agents.
Each agent has a different role:
• Strategist: Breaks down the problem and identifies the possible paths
• Researcher: Looks for relevant evidence and information needed to make the decision
• Analyst: Compares the options and evaluates the numbers, tradeoffs and expected outcomes.
• Risk Officer: Looks specifically for things that could go wrong.
• Devil’s Advocate: Attempts to destroy the leading recommendation. Not “here's another opinion”,Its job is to find the weaknesses in the argument.
Then everything goes to the final layer:
⚖️ The Judge
The Judge doesn't simply pick whichever answer sounds the most convincing.
It evaluates the arguments, it looks at evidence, it identifies contradictions, it considers uncertainty and then it produces a final verdict.
The interesting part is what happens when the agents disagree.
Imagine asking:
“I've been talking to someone for three months. I can't tell whether they're genuinely interested in me or whether I'm wasting my time. Don't reassure me, just tell me the truth.”
Instead of immediately generating a paragraph of relationship advice, Conclave can turn the problem into a deliberation.
• One agent looks at the behavioral evidence
• Another examines the long term implications
• Another looks for inconsistencies
• The Devil's Advocate tries to prove the initial conclusion wrong
• And the Judge weighs everything before giving you a verdict.
The final response isn't:
“Here are some things you could consider.”
It's structured like this:
CONCLAVE VERDICT
Recommendation: …
Confidence: …
Strongest evidence: …
Biggest contradiction: …
Biggest risk: …
What would change the verdict: …
Next action: …
And there's one principle I really wanted to build into it: Conclave must be willing to change its mind
If new evidence destroys the original recommendation, the system shouldn't defend its previous answer
It says:
“Our recommendation has changed.”
That, to me, is the difference between generating multiple AI responses and actually building a decision making system.
And the interface is just as important.
I built Conclave because I think the next generation of AI agents shouldn't just be answer machines.
They should be able to:
→ break problems apart
→ bring in specialized perspectives
→ challenge their own conclusions
→ reason about uncertainty
→ change their minds
→ and help humans make better decisions
And all of that can happen in a place we already know how to use - a text conversation
This is my entry for the Photon iMessage agent contest.
Built with @PhotonHQ
We’re giving $3000 to whoever builds the best iMessage agent:
Build on Photon → Post a demo & tag @PhotonHQ → Submit a typeform
💰$1,000 best use case
💰$1,000 most creative
💰$1,000 funniest / most viral
Deadline: Sep 10. Winners get invited to something special.. 👀
Happy shipping!
I’m hiring for @thecollectiv3e!
We’re looking for a few more fellows to join the team as we continue growing + launch in LA.
Come work with me, @maggiexgao, @FinnStaeblein, @Emhan022, and the rest of the team.
You must be based in SF or LA and be willing to consistently show up and help bring our events and community to life.
Here’s what you can expect:
- Meet an incredible network of people
- Work directly with founders + builders
- Help run the best events in SF & LA
- Build your personal brand
We’re looking for someone who’s ambitious, creative, and loves startups.
To apply, like + repost this and comment your email.
I'll send you the application directly.


























