August 25, 2026
TechCrunch's look inside OpenAI's push to build agents for "everything" -- paired with head of product Thibault Sottiaux's own comments that "the world seems to be ready" -- makes the company's near-term bet plain: agents aren't a feature anymore, they're the product. Sottiaux's job, reporting directly to Greg Brockman, is apparently to take agentic capability out of the coding niche and put it in front of ordinary consumers and knowledge workers. That's a much harder sell than it sounds. Developers will tolerate an agent that occasionally goes off the rails because they can read the logs and undo the damage. A marketing coordinator or a small-business owner won't extend that same patience, and won't necessarily know how to supervise an agent that's booking meetings, sending emails, or touching a CRM on their behalf.
The honest question TechCrunch's headline raises -- will everyone actually use them -- is the right one, and I don't think the answer is a foregone yes. Capability has clearly outpaced comfort. That gap is exactly why we've argued businesses should adopt agentic tools inside systems where the permissions and audit trail are visible and constrained by design, rather than handing a general-purpose assistant the keys to everything at once. If you're evaluating where agents actually belong in your operations, an internal tools and admin setup that scopes access narrowly is a far safer starting point than a do-everything assistant.
That gap between capability and comfort is already showing up in public. TechCrunch reports that early testers of Instinct's AI assistant are impressed by what it can do, but increasingly uneasy about the sweeping account access, broad terms of service, and autonomous action it requires to do it. This is the exact tension OpenAI is racing toward at much larger scale. An assistant that can act on your behalf is only as valuable as your willingness to let it act -- and that willingness evaporates fast the first time it does something you didn't authorize, or you simply can't tell what it's allowed to do. Security teams evaluating any AI agent, from Instinct to whatever OpenAI ships next, should be asking for the same things: scoped permissions, revocable access, and a real incident response plan, not a reassuring blog post. We've written before about why trust, not the underlying model, is the actual product vendors are selling right now, and Instinct is a live test of whether the market agrees.
Underneath the trust debate, the infrastructure for an agent-heavy future is being built fast. OpenAI's Jalapeño chip reportedly beat the current state of the art on SemiAnalysis's InferenceX benchmark, delivering more tokens per user and more throughput per kilowatt -- a signal OpenAI is serious about owning its inference stack rather than renting all of it. Meanwhile Accel-backed Keenable came out of stealth with a $26 million seed round to build a web search index specifically for AI agents, on the bet that agents need their own version of the web rather than scraping pages built for humans. Both moves say the same thing: the companies closest to this technology believe agents are going to need dramatically more compute and dramatically better information access than today's chatbots do. That's a reasonable bet on where the puck is going, but it's also a reminder that the businesses buying these tools are effectively subsidizing an arms race whose costs eventually show up in subscription pricing.
Hugging Face reportedly fielding acquisition offers around $13 billion is the most interesting values question of the week. The company's entire identity is built on being the neutral, community-run home for open-source models, and TechCrunch notes the founders themselves seem torn about whether a sale is compatible with that mission. A new owner -- especially a strategic one with its own model or cloud business -- could quietly reshape what gets hosted, promoted, or restricted on the platform millions of developers rely on as a default. If you depend on open, freely available models for your own products, this is worth watching closely, because the next steward of Hugging Face may not share the same incentives as the current one.
Which of these stories worries you more as a business buyer: an AI agent with too much access, or the infrastructure race that's about to make agents cheaper and more capable than ever?
Sources