{"slug": "sima-launches-agentic-development-environment-for-physical-ai", "title": "SiMa Launches Agentic Development Environment for Physical AI", "summary": "SiMa.ai launched Palette Neat, an agentic development environment for physical AI systems, enabling developers to design AI systems in plain English and reduce development time from weeks to days or hours. The tool, part of the Palette SDK, targets robots, drones, healthcare, and industrial automation by abstracting AI complexity and converting existing code to run on SiMa's Modalix MLSoC chip.", "body_md": "Edge AI chip company SiMa.ai has launched an agentic development environment designed for developers of physical AI systems working with the company’s [Modalix MLSoC chip](https://www.eetimes.com/sima-ais-second-gen-edge-ai-chip-goes-multi-modal/). Part of the company’s Palette SDK, Palette Neat, combines an execution library and an agentic workflow to speed up development time for robots, drones, healthcare, and industrial automation.\n\nDevelopers can design systems in plain English and develop in days, sometimes hours, SiMa CEO Krishna Rangasayee told EE Times.\n\n“To build a great silicon company, you need to have amazing software,” he said. “Our industry, unlike the data center market, is not really the most proficient in AI. The more we can abstract access to the technology and the more we can make it a push-button experience, the better the AI experience will scale.”\n\nSiMa aims to solve the embedded AI industry’s problems by allowing data from any sensor modality to run any AI model efficiently. A developer might prompt the agent with basic information about the system, its sensors, and constraints like accuracy and latency, and the agent can design a SiMa-based AI system accordingly.\n\n[View All](https://www.eetimes.com/category/sponsored-content/)\n\nSiMa’s Modalix hardware is a heterogeneous compute platform—an SoC with Arm CPUs and an in-house NPU. Physical AI is, in some ways, more complicated than cloud AI, since developers are dealing with “real-world challenges,” such as front-end logic, sensor integration, data parsing, different data types requiring pre- and post-processing, and more. Writing code for this can take 6-12 weeks in a typical customer team, Rangasayee said.\n\n“If your software is poor, which unfortunately is the norm across the industry, sometimes it takes months to iterate around this,” he said. “We’re dismantling this barrier with Neat.”\n\nThe new agentic workflow can condense weeks of effort into days or hours, drastically improving AI’s accessibility, Rangasayee said.\n\n“We have been so focused on numbers and speeds in the industry, but the maturity I see across customers is: all of that is great, but can you get me to production quickly?” he said. “Can you reduce the friction [to transition] from my investment in a GPU or DSP company and make it a SiMa product, and how quickly can you do it?”\n\nSiMa is engaged with all automotive OEMs through its [partnership with Synopsys](https://www.eetimes.com/synopsys-collaborates-with-sima-ai-on-automotive-ai-ip/), Rangasayee said. SiMa’s AI hardware IP and its toolchain are fully integrated into Synopsys’ architecture evaluation platform, simulation, and emulation environments to enable decisions, such as buying SiMa chiplets or licensing IP. Automotive customers have been impressed with Palette Neat’s capabilities, he said.\n\nSiMa shipped around 1,000 units of its production-qualified SoM last year, and around 20 customers have already engaged with Palette Neat in the run-up to general availability, Rangasayee said.\n\n**Code conversion**\n\nSiMa’s agents can convert existing code to run on SiMa hardware. Models such as Claude and Codex have been trained extensively on Nvidia Cuda kernels, and starting from code rather than from scratch can be effective, as it allows testing for equivalency, Manuel López Roldán, software product manager at SiMa, told EE Times.\n\n“It isn’t recompiling code, it’s more like what a human would do—analyze what the Cuda code is doing, then figure out the best way of doing that mathematical operation with SiMa,” he said.\n\nAgents can also create better code than humans can write in many cases, Rangasayee said, since they iteratively go through every option to optimize fully. This will go some way towards bridging Cuda’s perceived moat, he added.\n\n“Everybody now has access to the most expert engineering team you could get hold of, but you still stay at the abstraction layer, the simplicity, of just describing a problem in English and let the system really work for you,” he said.\n\nThere is a difference between using agents to solve the problem versus letting agents solve it while retaining control, Rangasayee said.\n\n“We are not reliant on some ethereal [thing] to go and do things which we worry about whether they did it right or not,” he said. “We know what we are doing. We are instructing it to do it to the degree and comfort and accuracy and testing that we want. And we have trained it with skills so that it always lives up to our quality standards.”\n\nAgentic AI has come a long way in the last six months, Rangasayee said, and is “robust,” with trust improving.\n\nSiMa’s existing Palette SDK will maintain access to the lower-level software toolchains for experienced developers, Rangasayee said.\n\n“People have a somewhat euphemistic view, that it’s somehow magic to work with Claude or Codex, and everything is magic,” he said. “But the secret sauce is we understand our architecture really well. We understand how to orchestrate things better than anybody else. And we are deploying agents to do the orchestration for us instead of human beings.”\n\n##### See also:\n\n[SiMa.ai’s Second-Gen Edge AI Chip Goes Multi-Modal](https://www.eetimes.com/sima-ais-second-gen-edge-ai-chip-goes-multi-modal/)\n\n[Synopsys Collaborates with SiMa.ai on Automotive AI IP](https://www.eetimes.com/synopsys-collaborates-with-sima-ai-on-automotive-ai-ip/)\n\n[Sima.ai CEO: Software Is The Company](https://www.eetimes.com/podcasts/sima-ai-ceo-software-is-the-company/)", "url": "https://wpnews.pro/news/sima-launches-agentic-development-environment-for-physical-ai", "canonical_source": "https://www.eetimes.com/sima-launches-agentic-development-environment-for-physical-ai/", "published_at": "2026-06-16 17:50:36+00:00", "updated_at": "2026-06-16 17:55:46.870203+00:00", "lang": "en", "topics": ["ai-tools", "ai-agents", "ai-infrastructure", "ai-chips", "ai-products"], "entities": ["SiMa.ai", "Modalix MLSoC", "Palette Neat", "Krishna Rangasayee", "Synopsys", "Claude", "Codex", "Nvidia Cuda"], "alternates": {"html": "https://wpnews.pro/news/sima-launches-agentic-development-environment-for-physical-ai", "markdown": "https://wpnews.pro/news/sima-launches-agentic-development-environment-for-physical-ai.md", "text": "https://wpnews.pro/news/sima-launches-agentic-development-environment-for-physical-ai.txt", "jsonld": "https://wpnews.pro/news/sima-launches-agentic-development-environment-for-physical-ai.jsonld"}}