September 24, 2026
TechCrunch reported that Lovable's annualized revenue has crossed $600 million, with co-founder Fabian Hedin saying apps built on the platform are pulling in nearly a billion monthly views combined. That's not a curiosity metric anymore -- that's a real product category with real usage. A year ago, 'vibe coding' sounded like a hobbyist trend: type a prompt, get a rough app, tinker until it works. Now it's generating enough revenue to rival established SaaS companies, and the apps people build with it are getting seen by actual users at scale.
Here's the part worth sitting with, though: a billion monthly views doesn't tell you how many of those apps are still standing in six months, or who's on the hook when one breaks in production. Prompt-to-app tools are great at getting something in front of users fast. They're a lot less proven at handling what happens after -- data migrations, permissions, integrations with the rest of a company's stack, the boring stuff that decides whether software survives past launch. We've made this case before when comparing Lovable-style prompt tools to building on ViibeStack and to Cursor: speed to a demo and durability as a business system are two different problems, and right now the vibe coding market is optimized almost entirely for the first one. Businesses evaluating these tools should ask not 'can it build this app' but 'can it build this app and still be maintainable a year from now.'
TechCrunch also reported that ChatGPT's mobile app is getting voice-based agentic features, letting Pro and Plus users trigger tasks in the Work tab by talking rather than typing. Pair that with Lovable's growth and a pattern emerges: the AI industry is racing away from chat-as-answer-machine and toward chat-as-assistant-that-acts. You don't type a prompt and read a response anymore -- you say what you want done and the software goes and does it.
That's convenient, and it's also exactly the kind of feature that sounds simple in a demo and gets complicated in a real workflow. Voice commands are great for quick, low-stakes tasks. They get riskier the moment an agent is booking something, sending something, or touching customer data on your behalf with no undo button in reach. Businesses adopting agentic tools -- whether it's ChatGPT's Work tab or something built into their own CRM or helpdesk -- should be asking what guardrails exist before letting an agent act autonomously, not after. It's worth noting OpenAI isn't alone in wrestling with this tension between capability and control -- Anthropic's own approach to its AI biology lab, discussed below, is a useful contrast in restraint.
Ando raised $20 million from Accel, Index Ventures and Emergence to build a Slack competitor designed from the ground up for humans and AI agents to work in the same channels, according to TechCrunch. It's a logical next step: if agents are going to act inside a company, they need a place to coordinate with the humans they're working alongside, and retrofitting that onto Slack is a real limitation. Whether Ando can out-execute Slack's distribution advantage is a separate question, but the bet itself reflects where the market thinks work is heading.
What's notable is the contrast with Anthropic, which told TechCrunch its biology lab has already made a significant discovery -- but pointedly has not let Claude operate the lab unsupervised. Humans are still in the loop by design. That's a useful data point for any business excited about agentic tools: even the companies building the most capable models are choosing to keep a human hand on high-stakes work. The lesson isn't 'don't use agents' -- it's 'match the level of autonomy to the level of consequence,' something worth thinking through before wiring agents into internal tools or customer-facing systems.
If your team is experimenting with vibe coding or agentic assistants right now, where are you drawing the line between letting AI move fast and keeping a human in the loop?
Sources