“A new AI assistant is rocketing across Silicon Valley.
Months after OpenClaw, the viral AI-powered assistant, captured the attention of the technology industry, a company called Instinct appears to be gaining traction among early-adopting techies.
The startup began testing Instinct in private beta in February and generated interest among venture capitalists, who are among its earliest users.
Its popularity surged earlier this month, as users began posting about what they saw as a highly capable AI assistant that worked fairly seamlessly, a goal technologists have considered a holy grail.
Instinct is raising a $250 million Series B at a $2.5 billion valuation in a round co-led by Index Ventures and Benchmark, founder Noah Shinn said. The steep price for the young startup reflects the continued frenzy surrounding AI investing.
Users of Instinct can call or text the AI bot and ask it to respond to emails, manage calendars, book a ride to the airport, arrange a handyman and more. Some users have reported using it to shop for homeowners insurance or order custom merchandise for a wedding.
"We saw someone buy a house on the platform. A lot of our younger users are using it to find apartment rentals," Shinn said. "It's a one-stop shop to do almost everything."
Silicon Valley has long been obsessed with using artificial intelligence to create personal assistants, but most of those ventures haven't succeeded. Now, advancements in AI are turning that pipe dream into a potential reality.
Instinct comes from Spear Street Technology led by Shinn, who was an early employee at Sierra, the customer service AI company founded by former Salesforce co-Chief Executive Bret Taylor. Shinn, 23 years old, dropped out of Northeastern University in 2023.
Since it was founded in 2025, Instinct has raised $350 million, including the new round, from investors including Conviction Partners and Greenoaks, Shinn said.
A range of apps are already helping people to unload menial tasks to AI. Instinct competes with ChatGPT Work or SpaceXAI's Grok Bot, which are growing more advanced at executing personal tasks.
Shinn said Instinct is "almost like having access to a superhuman operator."
Some early Instinct users have raised security and privacy concerns. To gain the most use from the assistant, users have to fork over access to sensitive personal data. Instinct's terms also allow it to train its AI models on user data.
Shinn said they are iterating fast and taking these security concerns seriously.
"We've made many product improvements that enable us to prevent the agent from taking certain actions that maybe some users of our early group who used an earlier version of the product might have experienced," he said.
Venture capitalists are pouring billions into AI assistants designed to automate daily life. Whether that be by writing and debugging software, coordinating family's schedules or picking a March Madness bracket, the bet is that autonomous AI will soon handle many of our everyday tasks.” [1]
Is AI Assistant Instinct with open weights? What open weight Chinese competition is better and runs locally on Mac Mini?
No, the viral personal assistant app Instinct is a closed, proprietary product, not open-weight.
Developed by Spear Street Technology and led by Noah Shinn, Instinct operates as a hosted, invite-only cloud service. It requires massive system permissions to execute real-world tasks like managing calendars and buying flights, which has actually triggered a wave of security and privacy concerns regarding how it stores and processes user data in the cloud. (Note: There is a separate coding-specific tool called "Instinct Next Edit Model" by Continue on Hugging Face, but it is entirely unrelated to the viral personal AI assistant).
If you are looking for better, open-weight Chinese alternatives that excel at agentic tasks, reasoning, and coding, you can achieve elite performance directly on consumer hardware.
The best open-weight Chinese models can be run locally on a Mac Mini (especially M4 or M4 Pro variants with unified memory) using the GGUF format via frameworks like Ollama or LM Studio:
1. GLM-5.3-Flash (Z.ai / Zhipu AI)
• The Background: Famously tested anonymously on OpenRouter as the stealth model "Ox Alpha," where it became a viral sensation for outperforming frontier-class models at a fraction of the cost. Its full weights are released under a permissive MIT license.
• Why it's better: It is a 320B parameter Mixture-of-Experts (MoE) model, but it activates only 18B parameters per token. It is natively multimodal (can process images/UI layouts) and approaches the agentic reasoning and coding capabilities of closed models like Claude Opus.]
• Running on Mac Mini: While the raw weights are massive (over 300GB), compressed Unsloth Dynamic GGUF or quantized IQ3 versions easily fit into the unified memory of a Mac Mini. Because only 18B parameters fire at a time, it runs remarkably fast on Apple Silicon.
2. Qwen 2.5 Series & Qwen 2.5-VL (Alibaba)
• The Background: Alibaba's open-weight flagship series is globally recognized for dominating open-source benchmarks in coding, mathematics, and structured data handling.
• Why it's better: If you want a visual agent capable of interacting with a computer or phone interface (similar to how Instinct acts), Qwen2.5-VL is specifically built to function as a visual agent that can dynamically direct tools.
• Running on Mac Mini:
o For a standard Mac Mini (16GB–24GB RAM), the Qwen2.5-7B-Instruct or 14B variants run at lightning speed locally.
o For a Mac Mini Pro (64GB+ RAM), you can run the highly advanced Qwen2.5-32B or a heavily quantized version of the flagship 72B model.
Local Setup Comparison for Mac Mini
Model Size Variant Best Mac Mini Fit Strengths
GLM-5.3-Flash 320B MoE (18B active) M4 / M4 Pro (Quantized GGUF) Exceptional agentic coding, stealth "Ox Alpha" reasoning
Qwen 2.5-VL 7B / 72B Base M4 (7B) or M4 Pro (72B quantized) Vision-language tasks, structured JSON, tool-use automation
Qwen 2.5 Coder 7B / 32B Base M4 (7B) to M4 Pro (32B) Elite local coding assistant, massive 128k context window
You need choosing the right quantization level (GGUF) for your specific Mac Mini's memory, and you need the Ollama terminal commands to get one of these models downloading and running locally.
1. AI Assistant Instinct Wows Silicon Valley. Clark, Kate. Wall Street Journal, Eastern edition; New York, N.Y.. 27 Aug 2026: B1.
Komentarų nėra:
Rašyti komentarą