We are entering an interesting phase in which AI manifests mostly as islands of intelligence with limited, unstructured connectivity. The transition to agents has been a major step in the adoption in solving complex problems. But that transition still lacks structure and standards, especially in how agents can communicate with complex entities such as organizations and tools. My focus has been on using AI to identify society’s pain points and unmet needs and to create solutions, technologies, and products to address them. Last year, we ran Spark 2025, which aimed to capture the needs of medical professionals and hospitals at price points that were globally acceptable. The outputs of that program were a market-confirmed and specialist-vetted “Request for Proposal” documents. The RFP captures in-depth requirements using SysML, informed by interviews as part of market engagement. These are then investigated using AI-assisted searches to establish the play in the market and identify the competitive landscape, size and IP whitespace. We then engage with specialists to further refine the requirements. The outputs of Spark are a SysML Block Diagram/Requirement Table and an RFP document.
In anticipation of this year’s Spark and Quest programs, I challenged the AI to take last year’s RFP for a 12-Lead ECG device and translate it into a real product. I used the technique used by Waymo in their self-driving cars, which, to me, is one of the most complex real-time applications of AI to date. Waymo uses a team of agents in its cars to achieve autonomy. One drives the car, the second synthesizes the reality as perceived, and the third, and the most important one, is the Critic. I created a team of agents with specialized skills and connected them to a number of tools whose APIs are exposed for agent use. This exercise was in anticipation of Spark and Quest 2026, where students will be developing products in conjunction with AI. Agents were from Claude, and CAD tools were KiCAD for schematic and PCB design. My role was simply that of an orchestrator. I challenged them to create an ECG device that costs less than $350 to ship. Its competitor prices range from $5000 to $20,000. It was like science fiction to see agents go to task and create an exquisite set of documents to start their work. They decomposed requirements and mapped them to both development and compliance documents, and created their own SysML block diagrams to work from. The key components were selected interactively, but the device’s size and scope were automatically determined.
| Architecture |
| Seven Interface Control Documents: |
| icd/ICD-PCB-FPGA-01 — AFE board ↔ FPGA interface |
| icd/ICD-FPGA-SW-01 — FPGA ↔ software interface |
| icd/ICD-PCB-CAD-01 — PCB ↔ enclosure interface |
| icd/ICD-PCB-PWR-01 — PCB power interface |
| icd/ICD-SW-MOBILE-01 — software ↔ mobile app interface |
| icd/ICD-SW-CLOUD-01 — software ↔ cloud interface |
| icd/ICD-CAD-USER-01 — enclosure ↔ operator interface |
| sysml/README_cameo_import.md — SysML import notes |
| PCB: pcb/design_narrative.md, pcb/esd_chain_addendum.md, pcb/SYSTEM_SIDE_CONNECTION_SPEC.md, pcb/afe_self_test_plan.md, plus pcb/kicad_bootstrap/ (directives + board outline spec) and pcb/ntx_ecg_12/ (ARCHIVED.md, design_intent.md, output summaries) |
| FPGA: fpga/design_narrative.md, fpga/memmap.md, fpga/filters/filter_design.md (note: superseded by the iter3-A STM32H7 pivot) |
| Software: sw/design_narrative.md, sw/soc_selection.md, sw/ai_architecture.md, sw/ble_gatt_spec.md, sw/cloud_arch.md, sw/mobile_arch.md, sw/recordings_db_schema.md, sw/firmware_architecture_iter3A.md, sw/security/stride.md |
| CAD: cad/design_narrative.md, cad/sealing.md, cad/drop_survival.md, cad/electrode_storage.md, cad/electrode_design.md, cad/thermal.md, CAD/environmental.md, cad/dfm.md, cad/models/enclosure_design.md, cad/models/enclosure_design_v11.md |
| Procurement: procurement/NTX-PATIENTCABLE-001.md |
In the end, the agents send the BOM and manufacturing details to companies like JLCPCB, which expose an MCP interface for PCB manufacturing and 3D printing. Agents can even drive AI model development by interfacing with companies like Hugging Face.
What is amazing is that it only took TEN days, a couple of hours in the evenings of agent orchestration, to go from an RFP document to the above. The final cost was less than $250. This compression of time is an exciting harbinger of what is possible for small teams looking to quickly prototype and create solutions.
There is a need to develop structured standards for exposing tools and businesses to agents to streamline the process and make it more robust. A company website should not only serve HTML but also expose an agentic API for every capability it wants to expose. The closest effort at scale in this realm that I know about is the Chinese Fifteenth Five-Year Plan.
This is the environment SPARK 2026 and QUEST 2026 students are stepping into, one where a small team with the right tools, the right framework, and the right problem can compress years of product development into weeks. That is what we are building together.
You can also read more about how to approach modern product development in my recent book.
I am excited for all those joining us in Spark 2026 and Quest 2026.