Documenting the milestones and breakthroughs on the road to AGI.
T-? to AGI
Joined November 2016
- Tweets409
- Following82
- Followers67.3K
- Likes551
Sundar Pichai says Google is launching the Gemini Decisions API in Google AI Studio, powered by the new Gemini Decision 1 model. It scores closed-set choices in a single pass rather than generating text token by token.
The API’s stated capability: "Get fast, calibrated decisions on up to 1000 questions at once across text, images, audio, video, and PDFs." xitter.twitterapi.xyz/sundarpichai/status/21…
Today we’re launching the Gemini Decisions API, powered by our new Gemini Decision 1 model in @GoogleAIStudio.
Instead of generating text token by token, it’s built for closed-set choices scoring up to 1,000 questions across text, images, audio, video + PDFs in a single pass. A few details:
- 3 decision types: yes/no, multiple choice (up to 255 options) + rubric scoring
- $0.10 / 1M input tokens with $0 billed for output tokens (~5x cheaper than Flash-Lite)
- leads on several benchmark evals with lowest calibration error
Google teams are already building with it, below is a look at Stitch Rapid designing a full product flow in seconds (1x speed)
Anthropic just dropped Claude Fable 5.2…but what does it really mean?
frontier labs want you burning $25/M tokens on every single turn of an agent run
even when 80% of the run is basic formatting and syntax checks
Rayline launched ARC today
an open-weight decision model that routes each task to the exact right engine and reasoning depth
stop paying frontier prices for baby tasks
🤝 Paid partnership
In 2015, I built an AI that detects early cancer.
Today, we’re introducing ARC by Rayline: the world’s first unified model for routing AI.
> 97.5% of Opus 5.5 quality
> Up to 1/10th of the cost
> Fully open-source
It picks the right model and the right amount of thinking for every step of your agent’s work.
Try it here and read our technical report below:
rayline.ai
Blue (YC S25) just launched a phone with a chief of staff built in.
It reads every notification, text, and email, handles what it can in the background, and drafts the rest for your OK - with full context across every app.
The team includes the person who led Google Assistant’s transition to LLMs and one who helped ship Apple Vision Pro. They looked at the current phone and decided the interface itself had to change.
10 people switched from iPhone. None wanted it back.
🤝 Paid partnership
We switched 10 people from iPhone to Blue. Weeks later, none of them want their iPhone back.
Introducing Blue: a phone you just talk to. Tell it what you need, and it gets done.
It has all the context, can use every single app, and even triages your notifications.
We built it for people who always have a little too much to do. The ones buried in messages, notifications, and to-dos.
If that's you, get yours today.
Backed by Y Combinator + 70 other VCs & angels.
heyblue.com
ByteDance founder Zhang Yiming has reportedly told his AI team that catching up with OpenAI and Anthropic in frontier AI could take until 2029 or 2030.
According to Bloomberg, he also acknowledged that ByteDance's Seed models have pretraining datasets several orders of magnitude smaller than those used by US rivals.
The company is preparing a Meta Muse-like AI assistant for launch within weeks.
ChatGPT and Codex are being forced to share one usage limit starting in December.
Grok Bot cut background task allowances in half.
Big AI is quietly rationing compute because running isolated agents burns too many tokens.
When Claude doesn't know what ChatGPT researched, and Codex doesn't know what Claude built, you're the unpaid context janitor shuttling prompts between tabs.
The frontier labs won't fix it because locked silos protect their recurring revenue.
Roma just launched to break it open: one shared task board where Claude, ChatGPT, and Grok work off the same list and execute offline from a quick voice note.
The era of babysitting disconnected bots is over.
🤝 Paid partnership
In 2014, I sold my first company Loom to Dropbox.
Today, I’m proud to announce Roma 2.0 (YC F26): the first to-do list for AI agents.
Everyone uses a different agent: Claude, Grok Bot, Muse…
Until now, none of them could work together.
Roma automatically creates tasks and hands them to the right agent.
Allowing you and your agents to work together, with shared context.
Try it here: roma.app
Plurai Simulation is now live and free.
Plugin for Claude Code, Codex, and Copilot that builds a digital twin straight from your repo.
The real agent runs unchanged in an isolated sandbox. Tools and world state are mocked (or kept live where it matters). Challenging personas walk every flow.
Auto evals check what the agent actually did to the world, not just the transcript.
Failures come with the proving conversation and a verified code-level fix. Production traces can be replayed from any step.
🤝 Paid partnership
Big day for us at @pluraiAI ! Plurai Simulation is live and free for every AI engineer.
AI agents look great in demos. Until real users break them.
So we built a world where your agent can fail, learn, and get fixed before your users ever see it.
→ Digital twin: Your real agent, its tools, and environment recreated from your code
→ Simulated users & tasks: Realistic users, tasks, and workflows that push your agent into the edge cases that matter
→ Auto evals: Check what your agent actually did, not just what it said
→ Verified fixes: Every failure comes with the run that exposed it and a fix you can verify
→ Replays: Turn any production failure into a simulation and replay it against every new version of your agent
All inside the coding agent you already use: @claudeai code, @OpenAICodexCli , or @GitHubCopilot .
Start simulating for free: simulation.plurai.ai/connect
If your team is actively building with AI and you’re not using Sunra Enterprise, you’re overpaying.
Team plans with up to 56% off the maker’s API prices on leading image, video, and creative models. Shared credit pool and admin controls in one workspace.
One of the cheapest ways for teams to make AI content at volume.
🤝 Paid partnership
n8n Labs ships Nod as a research preview: an open-source personal agent that runs locally with its own computer and browser.
It finishes multi-step jobs (call prep, invoice collection, subscription cancellations) across apps and websites, asks before acting outside your rules, and lets you bring your own model. 98% task completion in their tests.
🤝 Paid partnership
Nace AI’s Drex for Finance is a decision model built only for finance.
Give it a transaction, filing, invoice or statement + a question and the answer options. It returns a probability for each in under a second.
85.7% accuracy on 3,361 held-out finance decisions. +14.8 points over GPT-6 Luna.
Three-way match, AML flags, credit calls, filings QA - the yes/no and pick-one work that used to be manual, with a confidence score you can cut on.
Most models generate text and hope. Drex decides.
🤝 Paid partnership
Drex for Search plugs straight into Claude Code via MCP.
Five clicks: open a workspace, copy the command, paste it in. Claude can then search and cite documents already in Nace instead of you uploading them every time.
Document processing reads it, Drex for Finance makes the call, Search pulls the evidence.
🤝 Paid partnership
Try it on the Nace Console: console.nace.ai
Drex for Finance: nace.ai/drex
🤝 Paid partnership
Inkbox raises $5.3M led by Peak XV (with Guillermo Rauch, Henri Stern, Jonathan Frankle, Cristina Cordova, and Dan Levine) and launches hosted agents.
They already gave agents their own email, phone, iMessage, and WhatsApp. Now the agent runs right next to those channels: it wakes on inbound messages, replies with full context, then goes back to sleep.
The hard part of agent communication was never sending. It was catching the reply hours later.
🤝 Paid partnership
We raised $5.3M to give AI agents a computer to run on, with their own email, phone, iMessage, and WhatsApp
Led by @peakxvpartners, with @rauchg, @sternhenri, @jefrankle, @cjc, and @daniel_levine
Today we’re launching hosted agents on @inkbox_ai. Your agent runs next to its communication channels, so it never misses a reply. It sleeps when idle, so you don’t pay for downtime.
In the 4 months since our YC launch, we’ve processed millions of messages for AI agents.
Hedra launches Model Factory, bringing the recursive self-improvement loop to robotics, manufacturing, biology, and defense.
Researchers from Stanford, Berkeley, and UIUC (ex-NVIDIA, Meta Superintelligence Labs) post-train models alongside customer teams. A retrained H3 scored 34.9 on Google DeepMind’s Physics-IQ vs. 30.3 for H3 Max.
Flexible GPU sandboxes, autonomous training steered by Hedra researchers, and models that keep improving after deployment.
🤝 Paid partnership
Introducing Hedra Model Factory: an autonomous inference and training platform for building models for the real world.
The recursive self-improvement loop behind frontier research has been locked inside a few labs. We're opening it to teams in robotics, manufacturing, biology and defense.
Our researchers come from Stanford, NVIDIA and Meta Superintelligence Labs. Now we work alongside your team to post-train and optimize your models.
Graphon AI raises a $35M Series A six months after its seed to build persistent relational memory for foundation models.
The team (from NASA JPL, DeepMind, Meta, and Nvidia) constructs one self-supervised graph across multimodal data without predefined schemas. This preserves relationships, sequence, and context that vector RAG loses when it chunks data, enabling more accurate reasoning over large connected corpora at lower token and compute cost.
It ranks #1 on ATM-Bench and posts leading results on LoCoMo (94%) and DolphinBench (84%) while resetting the cost-accuracy frontier.
🤝 Paid partnership
After years working on robot memory at @NASAJPL, @GoogleDeepMind , @Meta and @nvidia, @clarkzhango and I kept asking: why does AI still forget?
Today, with @deepakmishraVC, we’re announcing @GraphonAI and our $35M Series A just 6 months after our seed.
• #1 on ATM-Bench, ATM-Bench Hard
• 94% on LoCoMo
• 84% on DolphinBench, 10x cheaper
• A new cost-accuracy frontier across 7 benchmarks
businessinsider.com/graphon-…
Stan acquires Typefully.
Typefully is the writing and publishing layer most people on X and LinkedIn already use. Stanley handles the harder part - figuring out what to say and turning it into a consistent content system for founders.
Writing and scheduling was never the real bottleneck. Distribution is becoming part of the job of building, and specialized tools are consolidating around that reality.
🤝 Paid partnership
Stan just acquired @typefully.
Typefully is where you write and publish. Stanley helps you work out what to say.
For six years, @linuz90 and @frankdilo obsessed over the craft. Typefully became the writing tool everyone recommends, and it got there almost entirely through word of mouth. We’re proud to carry that forward.
This is our biggest step yet in helping Founders create content that grows their business.
We can’t wait to show you what’s next.
Nuance Labs launches Nuplex, its full-duplex audiovisual foundation model, and Emma, the first persona built on it.
Emma listens and reacts at the same time, using visual cues for turn-taking and natural expressions rather than waiting and staring blankly. The team spent 18 months focused on the conversational dynamics people handle automatically.
🤝 Paid partnership
The first AI persona you’ll actually want to meet face-to-face.
Most AI personas stare blankly while you talk. Emma reacts while you’re still speaking.
After 18 months of research at @nuance_ai, this is Emma, built on Nuplex, our full-duplex audiovisual foundation model.
Demo is live if you want to try the full-duplex reaction.
Try Emma: app.nuancelabs.ai/try
🤝 Paid partnership
Volaren raised $1M from YC and Afore to build a personal investing agent that connects to brokerage accounts.
It explains why your portfolio moved (specific to your holdings), surfaces the next decision, and can place the trade once you approve. Available via text, web, or app.
Specialized agents moving from advice into execution.
🤝 Paid partnership
We raised $1M from YC & Afore to leave Goldman Sachs & build an agent that trades for you.
The top 1% gets a “money guy.”
Today we’re introducing Volaren so you can have one too.
Volaren is designed to plug into your brokerage accounts and explain why your portfolio moved, suggest what to do next, and trade it for you.
• Web, app, or text message
• Copy trade top investors/politicians
• Deep financial analysis
• Agents that reason like humans to trade your best ideas
Try it now:
volaren.com
Investing involves risk, and past performance does not guarantee future results. For informational purposes only, not a recommendation or investment advice. Other fees may apply, please review Volaren’s website and disclosures in full.
Readers added context they thought people might want to know
The announcement frames the founders as leaving jobs at Goldman Sachs after raising funds, but their prior post states the Goldman roles were summer internships before they incorporated; Volaren is in early beta on a waitlist with mostly simulated features.
x.com/akshat_mittal_…
volaren.ai
ycombinator.com/companies/vola…
athletics.uchicago.edu/sports/mens-te…
Skyfall AI launches Steve, the world’s first AI CEO harness for fully autonomous companies.
Powered by recursive self-improvement and world models, it plans via simulation, learns from failures, and coordinates specialist agents across growth, support, engineering, and operations. The team acquired three digital businesses and handed them over: one saw 2.3× MRR growth in under three months; another stabilized revenue and reduced churn after five years of decline.
They’re now allocating $2M to acquire five more startups for Steve to run.
🤝 Paid partnership
Introducing Steve: the World’s first AI CEO for fully Autonomous companies.
Powered by World Models, Recursive Self Learning, a Policy Engine and a team of specialist AI agents, it runs your business end-to-end 🔥.
Proof? Steve currently runs 3 businesses we acquired and they are all crushing it (numbers in thread 🧵)
We’re now launching the Steve Startups Program of $2M to acquire 5 additional startups for Steve to automate.
Join us in the Autonomous Enterprise revolution 🚀.