AI tools push toward agents
- Major vendors unveiled agent- and model-focused upgrades this week across design, models, and enterprise forecasts. - Canva launched 'Canva AI 2.0' as a conversational agent hub, Nvidia introduced Nemotron 3 Super, and Broadcom raised its AI revenue outlook. - Those moves point to deeper enterprise embedding of agentic workflows and larger-context models for production tasks ( ).
Artificial intelligence companies spent April pushing software toward “agents,” tools that take a goal in plain language and carry out several steps on their own. Canva, Nvidia and Broadcom each used the past six weeks to show a different piece of that shift. (canva.com, research.nvidia.com, broadcom.com) An agent is a chatbot with extra reach: it can plan, use tools, remember context and complete a task instead of only answering a prompt. That has become the pitch for office software, design platforms and the chips that run large models. (canva.com, research.nvidia.com) Canva said on April 15 that Canva AI 2.0 is its biggest product launch since the company started in 2013. The company described it as a conversational, agentic platform that can turn a prompt into layered, editable designs and added workflows for connectors, scheduling, web research, brand intelligence, Sheets AI and Canva Code 2.0. (canva.com) Canva said the product is powered by its Canva Design Model, which it called a foundation model built for design structure, layout and hierarchy rather than only text prediction. The company said Canva AI 2.0 began rolling out as a research preview on April 16 to the first 1 million users who find it on the homepage. (canva.com, businesswire.com) Nvidia made the model side of the same argument on March 10, when it released Nemotron 3 Super. Nvidia said the model uses 120 billion total parameters, activates 12 billion per token, supports up to a 1 million-token context window and is aimed at “agentic capabilities.” (research.nvidia.com, arxiv.org) Parameters are the adjustable weights inside a model, and a mixture-of-experts system works like a team where only some specialists are called on for each task. Nvidia said that design let Nemotron 3 Super deliver up to 2.2 times the inference throughput of GPT-OSS-120B on one benchmark setting while releasing pre-trained, post-trained and quantized checkpoints as open weights. (research.nvidia.com, arxiv.org) Broadcom supplied the spending signal behind those product launches in its March 4 fiscal first-quarter report. The company said AI revenue reached $8.4 billion, up 106% from a year earlier, and forecast AI semiconductor revenue of $10.7 billion for the second quarter. (broadcom.com) Broadcom Chief Executive Hock Tan said on the earnings call that the company had “line of sight” to more than $100 billion in AI chip revenue in 2027. CNBC reported Tan tied that outlook to demand from custom accelerator programs and named Anthropic, Meta and OpenAI as part of the customer momentum. (cnbc.com, broadcom.com) The common thread is that vendors are no longer selling only a smarter chatbot or a faster model. They are selling systems that can hold more context, call more tools and stay inside a workflow long enough to produce a campaign, a report, a spreadsheet action or a software task. (canva.com, research.nvidia.com) That leaves the next test in the same place it has been for the past year: whether companies move these tools from demos into daily work. This week’s launches showed the interface, the model and the hardware suppliers all betting that they will. (canva.com, research.nvidia.com, broadcom.com)
