22.10 Algorütm: Kuidas kiirendab AI riistvara arendust ja mis jääb inimese teha?

Summary

The episode examines how AI is changing custom hardware development, where it can accelerate engineering, and why physical testing and human judgment still matter.

AI in the hardware-development workflow

• Research and component selection: AI agents can compare far more parts and specifications than an engineer could review manually—potentially hundreds of options instead of a shortlist of three or four. Tasks that once took weeks of catalog searches and comparison tables can be condensed into a day, though components still need to be tested in practice.
• Schematics: The guest says current AI tools can produce readable, checked circuit schematics when given suitable instructions and documentation. His company moved from commercial electronics-design software to open-source tools partly because they were easier to extend and integrate with agents through MCP.
• PCB layout and compliance: AI is less reliable at deciding how a board should fit into its enclosure and connect to other physical components. Electromagnetic interference and regulatory compliance also remain difficult: individually compliant components can cause problems when combined. Hardware still needs real-world testing and, often, lab testing.
• Digital twins and simulation: In one example, Claude used measurements from a scientific publication to create a 3D digital model of a laboratory device. Its physics simulations led to the same design conclusions as the researchers. The guest sees this as a sign that AI’s engineering abilities are expanding beyond text and code, while noting that simulation tools and input data still have gaps.
• Prototyping: Faster, cheaper board manufacturing—especially in Asia—makes it practical to order custom development boards instead of wiring together multiple off-the-shelf modules. The guest describes sending a design to China in the morning, having it assembled within 24–48 hours, and receiving it about a week later. Shipping and customs still limit iteration speed outside manufacturing hubs.

What changes—and what does not

• Across the full hardware-development process, the guest estimates that development time has roughly halved. AI speeds up research and design, but physical production, delivery, component testing, and debugging remain.
• AI can make prototyping and exploring alternatives much cheaper. Engineers can generate several mechanical concepts and print prototypes that might previously have been too costly to test.
• Human expertise remains important for debugging, managing context, checking assumptions, and deciding what information to trust. The guest warns that AI may fill gaps with unsupported assumptions unless constraints and sources are clearly distinguished.
• Simple electronics projects may become accessible to non-specialists. Products that must work reliably for years—especially medical or business-critical devices—still require careful component choices, repairability planning, and attention to supply continuity and recertification.
• AI may help engineers become broader generalists, but the guest expects more people to work on physical-world problems, not fewer. Experimentation, user research, and real-world validation become more important as routine coding and documentation tasks shrink.
• There is no single definition of good user experience: it depends on the users and setting. An industrial automation dashboard and an interface for another audience have different needs.

Products, marketing, and business

• AI-generated product images, animations, and designs can look convincing even when there is no production-ready product behind them. This makes it harder for customers to distinguish credible products from “AI slop,” and established companies may benefit from their existing reputation.
• The guest is cautious about changing a website that already ranks well in search, since AI-generated or low-quality content could harm its visibility. He also notes that launching a new product without established authority or recognition is difficult.
• Faster development does not automatically mean lower prices. The guest says AI currently tends to improve the quality and value of the deliverable rather than reduce what customers pay.
• He avoids making customer communication sound machine-generated, arguing that a person’s natural writing style can help preserve trust and human contact.

Robotics

The guest is skeptical that humanoid robots are the best design for most jobs. Their many moving joints can make them complex and less durable; he favors purpose-built robots, such as robotic vacuums and lawnmowers, designed for specific tasks.

Reviews, guides, and tutorials

No product review, step-by-step guide, or tutorial is presented. The episode offers analysis and examples from the guest’s hardware-development work.

Main speakers and sources

• Tanel Ainla — founder of ScienceMosaic and hardware engineer; guest and primary source for the hardware-development examples.
• Priit Liivak and Martin Kapp — hosts of Algorütm.