On July 17, 2026, eSOL held its annual private forum, eSOL Technology Forum 2026, at the Tokyo Conference Center Shinagawa. Under the theme "The SDx Platform That Masters AI," six sessions examined what the AI-Defined Vehicle (AIDV) actually requires: platform-controlled data, concurrent development across mechanical, electrical and computing domains, accountability for foundational software in an era of AI coding agents, and the sensory brand character that end users still buy a car for. Eleven demo booths in the foyer showed the corresponding technology, from an AI agent harness for safe vehicle control to RISC-V, dual-OS industrial platforms, and an integrated SIL/HIL/real-vehicle validation environment.
Software-defined transformation (SDx), led by the software-defined vehicle (SDV), has passed peak hype. Generative AI is on its own curve, with foundation models near the peak and AI agents still emerging. eSOL's position at eTF 2026 was that both are heading to the same destination: technologies that will simply be taken for granted.
The bridge between them is the platform. SDx is, in the broad sense, platformization, and the platform is also what controls the data that makes AI useful. That is the argument the forum was built around, and every session approached it from a different angle: the OS vendor's view, the Tier-1's view, the software-history view, the product-management view, and the end user's view.
Session Summaries
| The SDx Platform That Masters AI: Untangling the Truth of the AI-Defined Vehicle from the Perspective of an OS Vendor Masaki Gondo, President, CEO & CTO, eSOL |
|
Learn more
SDx currently sits in the trough of disillusionment on the hype cycle and is likely to enter the slope of enlightenment within about three years. Generative AI is on a parallel but offset track, with peak, emerging and disillusionment phases coexisting across different technologies. What the two share is inevitability, because both will become ordinary infrastructure. Gondo structured the keynote around four questions: (1) What is actually happening right now in manufacturing and product creation? (2) What is the relationship between the platform and AI within SDx? (3) What new world of application development do the two together create? (4) What does concurrent development look like when a single platform spans mechanical, electrical and computing domains? He closed with eSOL's own direction for delivering on each. Key takeaways - AI's decisive resource is data, and the platform is what controls data. Controlling SDx means controlling the platform. - AI is probabilistic and never deterministic, so it must not own critical real-time paths. The platform acts as a harness, with the API refusing unsafe requests. - SDV enables AIDV, which enables Customer Defined Vehicle: users generate functions on the spot, and their usage data flows back into development. |
| The Essence of "Software Defined" and Its Key Technologies: Software-First Innovation in Vehicle Development in the SDV Era, and Beyond to AI-Defined Masashige Mizuyama, Representative Director, Executive Vice President, CTO / CISO / in charge of Intellectual Property, Panasonic Automotive Systems Co., Ltd. |
|
Learn more
SDV is a genuine game changer, because it shifts the axis of automotive value and competition from hardware to software. Whoever can evolve software fastest wins, and the contest moves from maximizing code output to product discovery: how fast you can fail, improve and arrive at the product actually worth building. Mizuyama grounded this in the mobile phone industry of the 2000s, where software-defined-ization drove horizontal specialization that could not be resisted, and handset makers who hesitated over architecture renewal and ecosystem building declined. He then walked through the transformations SDV demands, Panasonic Automotive Systems' own answers to them, and what lies beyond SDV in AI-Defined. He illustrated each point with the company's initiatives. Key takeaways - Once a category goes software defined, ecosystem influence jumps to another level. The contest shifts from differentiating features to network effects, and losing a chokepoint to someone else means commoditization. - Software first means software defines the hardware spec, and VirtIO is the technology that makes it possible. PAS open-sourced it in 2018 and led its standardization, winning industry-wide endorsement in February 2026 and already shipping in the new Mazda CX-5. - AI-Defined is a necessity, not a chatbot. User requirement diversity runs orders of magnitude beyond what can realistically be specified and tested, so AI must handle meaning and context directly, and the developer's role shifts to designing training data, evaluation metrics and guardrails. |
| From Assembly to Agents: What History Tells Us About the Future of Software Nikos Michalakis, Founder, Red Vest Mindset; Co-founder & CTO, ActBI Inc. |
|
Learn more
Every generation of engineers believed they had mastered software. Every generation was then abstracted away by the next layer. Michalakis placed AI in exactly that 70-year lineage: not a disruption, but the next abstraction in a very old pattern that runs from assembly to compilers, object orientation, composition, declarative infrastructure, and now agents. Today about 30% of new code at Google and Microsoft is AI-written and agent teams open five times more PRs, yet AI-assisted code carries 1.7 times more bugs and only 4% of developers fully trust the output. The consequence he drew is economic as much as technical. When the cost of building software falls by a factor of a thousand, the assumption behind "one product for everyone" may already be obsolete, because software can be shaped per user, per context, per moment. He closed on what is missing: the tooling and process gap that the agentic era urgently needs to close before that future is safely buildable. Key takeaways - AI is the next abstraction layer in a continuous 70-year progression, not a break from it. Each layer traded performance for productivity, and that bet has paid off every time. - A 1000x drop in build cost undermines the one-size-fits-all product model, making per-customer variation something to architect for rather than avoid. - The binding constraint is no longer generation. It is tooling: specification validators, acceptance verifiers, constraint checkers, agent trace debuggers, and source-level optimizers against slop. |
| Is the Proprietary Software Business Dead? Beyond Abstraction, Who Takes Responsibility for Foundational Software? Soichiro Matsumoto, General Manager, Engineering Division / Product Manager, Product Management Department, Business Management Division, eSOL |
|
Learn more
As AI coding agents improve, producing working code from a public specification has become dramatically easier, which raises a real question inside many organizations: if AI can build foundational software, are foundational software vendors still needed? Matsumoto's answer was deliberately split. Yes, AI can build foundational software that runs. But being able to build something means more than running to spec. It means being able to fix it, improve it, and explain it. He drew the parallel with the OSS era, when the same obituary was written for software vendors and proved wrong for the same reason. What a customer needs is a co-owner who translates the problem and stays to solve it to the end, rather than an implementer who does the wrong thing correctly, illustrated with a throughput complaint answered by switching to polling and shared-memory IPC. His closing question, what customers can divide with eSOL and how, is answered by white-box partnership: lifecycle ownership of the foundational software held jointly, on the premise that responsibilities and concerns are separated but visible. Key takeaways - AI-generated foundational software is feasible to produce and hard to own. Running to spec is not the same as being able to fix, improve and explain it. - The durable value shifts from possessing code to being a co-owner of the problem through the full product lifecycle. The platform is the harness for that lifecycle. - "Buy vs. build" should be reframed as a third option: white-box partnership alongside make and buy. - What dies is any business whose only value is holding code. What survives is one that keeps a responsible foundation together with its customers. |
| In the AIDV/SDV Era, What Do Users Think About When Choosing a Car, and What Do They Want? OEM Brand Character and the Business Model Beyond It Akira Takahashi, Juror, Japan Car of the Year (COTY); Secretary General, Japan Motorsports Press Association (JMS); Archive WG, Motorsport Division Committee, Society of Automotive Engineers of Japan (JSAE); Member, Automotive Journalists Association of Japan (AJAJ) |
|
Learn more
Among automotive journalists, the common reading is that cars will eventually become "like smartphones," and that this will create new value. But there is a persistent worry that as software-defined architecture advances, the pleasure of driving thins out. Takahashi opened by noting that outreach asking journalists to understand SDV has not really reached them. He turned the question around: what is it that people actually find appealing in a car? Design, drivability, a private space for oneself or one's family, the joy of controlling a machine that moves exactly as intended. These are sensory, multimodal qualities, and AIDV/SDV should be raising the floor in exactly that domain. He illustrated this with the next-generation Rolls-Royce Nightingale concept, where technology shifts from adding sound to selecting it, hearing is converted into sight through some ten thousand points of light, and the sensory experience itself becomes the product. What is sold there is not a car but the design process for an experience. He then contrasted brand character across markets, and asked whether the character OEMs pursue is truly aligned with what users want. Key takeaways - The purchase decision stays multimodal and sensory, and value is shifting from owning to experiencing. Software should quantify and amplify those qualities, not flatten them. - OEM-defined brand character risks drifting from what users want. Renault and Citroën each drive like their brand in the same MPV category, while Japanese design tends to follow whatever succeeds. - IVI and dynamic performance never appear on a spec sheet, so the contest is over sensory value, and today it cannot be monetized beyond vehicle price. That keeps the old build-and-sell model in place. |
| The Power of the Platform in the AIDV Era: Execution and Development Foundations That Keep Creating Value Yuta Nakauchi, General Manager, Product Management Department, Business Management Division, eSOL |
|
Learn more
The closing session tied the day together. Nakauchi revisited all five talks, namely the essence of AIDV/SDV, software-first and open ecosystems, agents as the next abstraction layer, the responsibility demanded of foundational software, and the shift in user value, and drew a single conclusion from them: an intelligent AI-equipped vehicle is only one facet of AIDV. The essence is the mechanism that connects the world of those who build things with the world of those who use them, and keeps raising user value over time. What connects those two worlds is the SDx Platform. From there he addressed the practical question the audience came for. An SDx Platform is made of three forces: the power to protect, meaning the runtime that controls non-deterministic AI safely through APIs; the power to turn, meaning the DevOps foundation that keeps AI-assisted development cycling; and the power to connect, which binds the two through shared models, APIs and data. The eSOL AI Mobility Sandbox was presented as the concrete showcase, with its value creation loop running from a natural-language user request, through DVE+U vehicle APIs, into logging and analysis, and back into improving the agent, the API and the platform design. Key takeaways - Execution platform and development platform must evolve as one, and a third force, the power to connect them through common models, APIs and data, is what makes them a platform at all. - AIDV is a mechanism for continuously evolving user value by linking AI, data and software, not a single capability shipped in a vehicle. - The AI Mobility Sandbox is eSOL's proposed approach, shown as a live demo and offered as a starting point for discussion rather than a finished answer. |
Eleven Demos in the Foyer
Alongside the sessions, eleven demos ran in the venue foyer all day. Every break turned into a crowd, and the networking session ran well past its scheduled end, which is the best feedback a demo team can get.
Here is what was on the floor:
| Demo | What it covered | |
| 1 | eSOL AI Mobility Sandbox | An AI agent driving a vehicle through natural language, using DVE+U model APIs |
| 2 | eMCOS®, Scalable Real-Time OS Platform | Multikernel RTOS scaling from MCU to HPC class silicon |
| 3 | eSOL AUTOSAR Solutions | Platform development support and service case studies |
| 4 | eSOL eXRP™, Real-Time 3D Engine | Simulation, digital twin and HMI on one real-time 3D foundation |
| 5 | eSOL ROS Engineering & Product Co-development | Embedded ROS 2 services and NEDO project results on real hardware |
| 6 | eSOL RTOS/Linux Dual-OS Industrial Platform | eMCOS Hypervisor with Cybertrust's EMLinux for mixed-criticality systems |
| 7 | eSOL TRINITY Security Solutions | CRA readiness consulting and a test platform |
| 8 | KMC Development & Integrated System Analysis Tools | SOLID, PARTNER-Jet3, and system-wide measurement on one timeline |
| 9 | Infineon RISC-V × Synopsys × eSOL | eMCOS and AUBIST on a RISC-V MCU, sensor to visualization |
| 10 | NXP × eSOL | eXRP and EMLinux on i.MX95, plus the eMCOS port for S32N55 |
| 11 | dSPACE × eSOL | SIL, HIL and real-vehicle validation unified for CI/CT/CD |
Materials you can download
We have put together English handouts for five of the eleven demos. These are the same panels visitors read at the booths, so you get the architecture diagrams and configuration details, not just a product blurb.
| Demo | What's in the handout |
| eSOL AI Mobility Sandbox | Enabling AIDV: AI agent overview, the DVE+U model, and the full system configuration diagram |
| eMCOS®, Scalable Real-Time OS Platform | Scalable Real-Time OS Platform: multikernel design, safety standards, and the eMCOS SDK |
| eXRP® enterprise Godot | Real-time 3D engine for industrial use: features and application examples |
| eSOL RTOS/Linux Dual-OS Industrial Platform | eMCOS Hypervisor × EMLinux configuration for mixed-criticality systems |
Session video and presentation materials
You can also watch "From Assembly to Agents: What History Tells Us About the Future of Software", the English keynote by Nikos Michalakis (Founder, Red Vest Mindset; Co-founder & CTO, ActBI Inc.).
After completing the form and you will be redirected to the download page with everything on it.
Want to know more?
We hope this recap gave you a good overview of what eTF 2026 was like. This article is based on the presentations from the event and may be updated as needed.
E.O.
Marketing Communications Team