A Nucleo board with an expansion shield during bench testing

A practical selection guide to the STM32 Nucleo family: the three form factors, what the order code tells you, what the on-board ST-LINK actually does, and how Nucleo compares to Discovery kits and to Blue Pill and Black Pill boards.

STM32 Nucleo Boards: How to Choose the Right One

An STM32F103C8T6 Blue Pill wired into a prototype.

A working reference for the STM32F103 line: the four density classes and what each gives you, why USB and CAN can't run together, the LQFP48 pin functions, how the Blue Pill is wired, and how to spot the clones on cheap boards.

STM32F103: Density, Packages, and the Blue Pill

<h2 id="isPasted">What Are AI Agents in Engineering? From Single Agents to Scalable Agentic Systems  </h2>AI agents in engineering are rapidly moving from experimental pilots to practical systems that execute real engineering work. Unlike copilot AI tools that generate text or graphics with mainly office or project management tools, AI agents in engineering are designed to operate directly within engineering tool workflows.  They can connect language models with domain-specific context, structured data, and specialized tools such as CAD, simulation, and costing software. The result is a new class of digital coworkers that can perform multi-step engineering tasks with human oversight. Large Language Model + Context + Tools = Agent <ul><li>Large Language Model: The "brain" for logic, reasoning, and understanding instructions. ‍</li><li>Context: the specific goal, role or job the a is meant to perform‍adhering to engineering requirements, regulations, and constraints.‍</li><li>Tools: External systems, such as PDF reader, Excel, or even specialized engineering tools and workflows. </li></ul><img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1789128556853-697c7fa7b0dc7199abe0a4b2_Synera+Agent+inside-20260128-131802-p-2000.png" style="width:100%px" class="fr-fic fr-dib" />‍By equipping an LLM with context and tools, its function transforms. Instead of merely providing theoretical text-based answers, an agent can interact with engineering systems, manipulate them directly, and deliver solutions that an engineer can rely upon. It moves from being a conversationalist to being a collaborator. However, this single-agent architecture soon encounters its own limitations. The sheer volume of information and contextual complexity inherent in engineering is something that a single agent is not fully capable to handle. This created the need for a more sophisticated, collaborative framework to manage the vast web of engineering data and processes.  This realization leads directly from the concept of a single agent to the necessity of a multi-agent system (MAS), designed specifically to orchestrate AI capabilities at an enterprise scale. <h2>Multi-Agent AI Systems for Engineering Workflows </h2>Agentic AI systems that specialize in engineering, like those implemented with Synera, represent the state-of-the-art solution for orchestrating teams of AI agents in complex domains. These teams consist of specialized agents, like a CAD agent, an FEM agent, and a cost agent, that work together across multiple steps in a product creation process.  ‍<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1789128574605-697b20096e4c332400fc62b7_Synera-Multi-Agent-System-RFQ.webp" style="width:100%px" class="fr-fic fr-dib" />This multi-agent approach was born from the practical need to manage the vast combinations of context and tools that can overwhelm a single AI agent. Instead of a user interacting with one agent connected to a set of tools, Synera users engage with a supervisor agent, that in turn orchestrates a team of specialized agents, digital engineers, where specialists (like a"CAD designer" or an “FEA expert") work together to deliver much higher quality on detailed tasks.Each digital engineer possesses domain-specific engineering knowledge, access to specific tools, like Moldflow, Fusion, HyperMesh or Siemens NX and clear process workflows to ensure quality results engineers can trust.  This architecture developed by Synera allows the team of agents to deliver tested product designs, while the workflows act as guardrails to prevent hallucinations, ultimately ensuring results engineers can trust. This framework helps AI advance from simply being able to "talk and reason" to being able to "act and analyze" delivering fast product development results in an efficient collaboration and work split with human engineers. Human engineers are the creative and innovative minds and get answers to their ideas very fast from agentic coworkers. As an example, see how AI agents work for NASA in this short video: ‍<iframe width="640" height="360" src="https://www.youtube.com/embed/8YRkwy6Vg90?&amp;wmode=opaque&amp;rel=0&amp;enablejsapi=1&amp;origin=https://www.wevolver.com" frameborder="0" class="fr-draggable">     </iframe> <h2>How AI Agents Are Used by Engineering Teams </h2>Engineering teams use AI agents as digital coworkers that combine the reasoning capabilities of large language models with an orchestration layer and direct access to rule-based workflows and engineering tools. The agents do not replace engineers. Instead, they learn from engineering decisions, enforce consistency, and carry context across engineering tools and teams, relying on humans when goals, quality criteria or judgment are required. <h3>CAD automation </h3>How AI agents work: AI agents generate and modify CAD models by directly interacting with industry-standard CAD tools, while engineers define requirements and approve results. <ul><li>Engineers describe requirements in natural language, such as functional constraints or design intent. </li><li>Specialized CAD agents translate these inputs into structured actions and execute them directly in tools such as CATIA, Creo, or NX. </li><li>For parametric designs, agents adjust parameters and regenerate geometry through workflows, like those built in <a target="_blank" href="https://www.synera.io/?__hstc=126354002.4583f3288c43f88541c46c1c1a7d17f0.1782147122075.1789117915145.1789125136034.211&amp;__hssc=126354002.5.1789125136034&amp;__hsfp=fb60cbe419693564f26d59c0c22e148d">Synera</a>, ensuring consistency without manual rework. </li><li>Throughout the process, agents apply institutional rules such as tolerances and standards, and request human review when conflicts or ambiguities arise. </li></ul>Human oversight:  Engineers validate geometry, approve changes, and intervene when design tradeoffs require expert judgment. <hr /><h3>Design-to-cost optimization </h3>How AI agents work:  A team of agents evaluate cost implications of design decisions early, using costing tools rather than estimates. <ul><li>Costing agents analyze CAD and BOM data and connect directly to expert software such as Facton or custom databases to calculate physically reasonable costs.</li><li>Engineers ask “what-if” questions, such as how cost changes with production volume or material choice. </li><li>Agents automatically run scenarios, compare alternatives, and surface tradeoffs for review. </li><li>When lower-cost options are feasible, agents coordinate with CAD and process agents to propose design alternatives. </li></ul>Human oversight:  Engineers and cost specialists review assumptions, approve scenarios, and decide which tradeoffs align with business and performance goals. <hr /><h3>Simulation and validation </h3>How AI agents work:  Simulation agents automate setup and execution while preserving engineering intent and physical correctness. <ul><li>Agents interpret design contexts to define load cases, boundary conditions, materials, and solver settings in FEA and CFD tools. </li><li>For computationally intensive simulations, agents use reduced-order models to provide fast, directional insight before full validation runs. </li><li>Multi-agent workflows execute large-scale validation, such as overnight simulations across many design variants. </li><li>Results are summarized and flagged for anomalies, with traceability back to assumptions and inputs. </li></ul>Human oversight:  Simulation experts review results, validate assumptions, and decide when full-fidelity analysis is required. <hr /><h3>RFQ Evaluation and Selecting Suppliers </h3>How AI agents work:  Agents coordinate data and analysis across departments to accelerate the RFQ process without losing engineering precision. <ul><li>A supervisor agent orchestrates inputs from a team of agents: sales, engineering, assembly, and purchasing. </li><li>Agents extract key requirements from large customer documents, identify inconsistencies, and translate them into structured engineering data. </li><li>Sourcing agents evaluate supplier strategies using cost, risk, and feasibility criteria. </li><li>The team of agents assemble RFQ packages and initial estimates automatically, surfacing open questions for human input. </li></ul>Human oversight:  Engineers and sourcing teams review requirements, validate assumptions, and make final supplier and pricing decisions. ‍<h2>Agentic AI opportunities and challenges in engineering </h2>Engineering is a unique challenge that can’t be easily solved by any AI agent solution. Engineering relies on structural knowledge and accurate physics, such as CAD data and simulation results, for which LLMs and generalized agents are not primarily designed.  Robust and seamless tool integration across the product development process is another key to overcoming a fundamental hurdle to agentic AI in engineering. Equally important is protecting the IP of manufacturers. Cyber security is a key reason why Synera AI agents are deployed via on-premises infrastructure and not via the cloud. According to my understanding of the Gartner® report Manufacturing Predicts 2026: AI Agents, Digital Twins and the Race to Autonomous Operations, while AI agents are advancing toward autonomous operations, adoption in the manufacturing sector lags due to these specific barriers: <ul><li>Technology integration challenges </li><li>The high stakes of safety-critical environments </li><li>Limited AI readiness within organizations </li><li>Persistent issues with data quality and availability </li><li>Heightened cybersecurity concerns </li></ul>Agentic AI solutions for engineering must successfully navigate these challenges – technology and organizational change need to go hand-in-hand. Those that do, will define the pace at which engineering-heavy enterprises move from agentic AI pilots to an established AI engineering operating model.   <h3>Navigating the 40% Cost Escalation </h3>Additionally, there will be an increasingly prohibitive cost factor catalyzed by this autonomous shift. Gartner predicts: “By 2029, the annual cost for manufacturing core systems such as product development, manufacturing execution and PLM software and related services will rise by 40%."Gartner notes that, "Deeper lock-in with incumbent enterprise-wide platform providers encourages users to seek lower-cost technology providers offering high cloud-native AI value. These new-generation technology companies will gain traction, challenging incumbents and reshaping competitive market dynamics.”  Success requires solutions that move beyond fragmented pilots, overcome organizational barriers to adoption, and build a robust automation foundation where agents and humans share a flexible automation platform that enables switching between different CAD, CAE, meshing or solver, and PLM applications very easily to choose the most suitable ones and enable smooth tool transitions should cost-saving become a factor.  <h3>Quantifying AI adoption and the competitive imperative for business executives </h3>To make informed investment decisions and set realistic technology roadmaps, it is critical to quantify the rate of AI's progress.  The number of human-guided multi-agent systems in engineering is increasing rapidly today. These have already improved efficiency and lead time. As the next phase of agentic AI emerges, the application of semi-autonomous agents in key use cases has impressive potential to further solve engineering pressure to do more with less.  A strategic planning assumption from Gartner predicts that “by 2030, semi-autonomous AI agents will orchestrate 10% of key production operations, quality, and maintenance use cases.” This represents a significant increase from just 2% today, though it is emphasized that humans will retain final approval for all critical decisions. You can download the Gartner report here.The fact that the tool layers for engineering are pretty much fixed means that the true competitive differentiator is no longer access to the latest AI model, but the organizational readiness and specialized agentic engineering solutions that can connect to the engineering tool layer to deploy agents effectively and equip them with powerful tools.  The competitive imperative for every leader is therefore clear: shift focus from evaluating the next AI model to architecting the operational context: the integrated data, tools, and workflows that agentic AI systems require. The organizations that build this foundation today will be the ones that are first to capture the value of AI tomorrow.

From Single Models to Scalable Agentic Systems

AI Agents for Engineering

Multilayer ceramic capacitors (MLCCs) are built into the fabric of modern electronics. They have been a default choice because they offer compact size, low ESR, low ESL, and strong high-frequency performance. In dense embedded, industrial, automotive, communications, and computing designs, they remain essential for local decoupling, filtering, and many power rail functions. Capacitor selection, however, has become a more strategic design decision.Modern hardware is becoming denser, power profiles are less predictable, and production planning has to account for availability, lifecycle status, and validated alternatives earlier in the design process. As a result, engineers are looking more carefully at where MLCCs are the right choice, where other capacitor technologies perform better, and how a mixed capacitor strategy can make a design easier to validate, source, and scale.This is where alternatives to MLCCs come to the fore. They are part of a broader shift in how engineers think about capacitance, from a nominal value on a bill of materials to a function that has to hold up under real electrical, thermal, mechanical, and supply conditions.<h2>The Design Question Behind MLCC Selection</h2>The primary issue comes in the shape of effective capacitance in circuit conditions. In many designs, the capacitance available in operation can differ significantly from the value printed on the datasheet.High-capacitance ceramic capacitors, especially those using high-permittivity dielectrics, can lose capacitance as DC voltage is applied. Temperature, frequency, ageing, board stress, package size, and mounting conditions can also affect how the capacitor behaves once it is placed in the circuit.This matters most when the MLCC is expected to provide bulk capacitance, support a converter output, stabilize a rail, or help manage transient current. If the effective capacitance drops under operating voltage, the circuit may have less margin than the schematic suggests. The issue becomes more visible in compact designs where engineers try to meet capacitance, voltage, and footprint requirements at the same time.Stable ceramic dielectrics can offer better capacitance stability, although usually at lower capacitance density. High-capacitance MLCCs provide more capacitance in smaller packages, although their behaviour must be checked under the conditions the circuit will actually see.Bottom line, the important question is how much usable capacitance the design receives across voltage, temperature, frequency, and time.<h2>Where Alternative Technologies Start to Make Sense</h2>Alternative capacitor technologies become relevant when the circuit needs properties that are difficult to achieve with MLCCs alone. The decision depends on the role of the capacitor, rather than on capacitance value alone.Polymer aluminium and polymer tantalum capacitors can support low ESR, stable bulk capacitance, and power rail smoothing in compact designs. They are often considered where engineers need more stable capacitance under bias, stronger ripple current handling, or reduced dependence on a large number of parallel MLCCs.Conventional tantalum and niobium oxide capacitors can be useful in compact, high-capacitance positions, especially where board area is limited and capacitance density matters. Their use requires attention to voltage derating, surge current, failure behaviour, operating environment, and qualification requirements.Aluminium electrolytic capacitors remain relevant where bulk energy storage, voltage handling, and ripple current capability matter more than board density. Film capacitors can support applications that need stable electrical behavior, voltage tolerance, low losses, or long-term performance, although their size and mounting requirements can limit their use in dense assemblies.The practical value of these alternatives is that they give engineers more ways to match the capacitor technology to the circuit function. A local high-frequency decoupling position, a DC-DC converter output, a hold-up rail, and a high-ripple power stage do very different jobs. Treating them as equivalent capacitor placements can lead to weak design assumptions.<h2>Alternatives Require Circuit-Level Validation</h2>A capacitor alternative cannot be selected only by matching capacitance, voltage rating, and package size. Changing capacitor technology can affect various properties, including impedance, ESR, ESL, leakage current, ripple current, self-heating, acoustic behaviour, transient response, and control loop stability.This is particularly important in power conversion circuits. Output capacitors are part of the control loop, and changes in ESR or capacitance can affect stability, transient response, and ripple. In high-speed digital designs, the impedance profile of the decoupling network matters across frequency, so replacing ceramic capacitance with another technology may require changes to the full capacitor network rather than a simple part substitution.Thermal behaviour also needs attention. A capacitor with suitable electrical ratings may still perform poorly if ripple current, ambient temperature, or board placement increases self-heating. Leakage current can also matter in battery-powered or low-standby-power designs, where small current paths become meaningful over time.The safest alternative is the one that fits the function the original capacitor performs in the circuit. For that reason, capacitor substitution should be treated as a circuit decision and validated under the conditions that define the product’s real use.<h2>Designing a Mixed Capacitor Strategy</h2>Many modern designs benefit from combining capacitor technologies according to their strengths. For example:<ul><li>MLCCs can remain the primary choice for high-frequency local decoupling near ICs. </li><li>Stable ceramic dielectrics can support precision or low-drift positions. </li><li>Polymer capacitors can provide bulk capacitance and power rail support. </li><li>Tantalum or niobium oxide capacitors can serve compact high-capacitance roles when their application limits are properly addressed. </li><li>Aluminium electrolytic and film capacitors can support positions where energy storage, voltage handling, ripple current, or long-term stability carry more weight than board density.</li></ul>This mixed approach can also reduce the risk created by over-concentrating capacitance in a narrow MLCC specification. A design that depends heavily on a specific capacitance value, dielectric, package size, and voltage rating may have limited room to adapt during production. Conversely, a design that distributes capacitance roles across suitable technologies can create more flexibility in sourcing and validation.This becomes especially important in industrial, automotive, medical, communications, and long-life embedded systems, where products must remain available and serviceable for years. A capacitor network that performs well during prototype testing also has to remain practical across production, field support, and future revisions.<h2>Supply Resilience Starts at the Circuit Level</h2>Capacitor strategy is also part of Bill of Materials (BOM) resilience. Some capacitor positions are easy to substitute because they have broad electrical tolerance and many qualified options. Others are tied closely to converter stability, high-frequency impedance, thermal limits, safety requirements, or product certification.Knowing the difference helps engineering and sourcing teams focus attention where it matters. Critical capacitor positions should be reviewed early for availability, lifecycle status, alternative technologies, package options, and validation requirements. This gives teams more room to respond if a specific MLCC value, dielectric, or package becomes difficult to source.Win Source’s role is to help engineering teams evaluate component availability, identify alternative capacitor options, understand lifecycle risk, and support continuity from prototype to production. For capacitor networks, that support is most useful when it is connected to the circuit function, rather than limited to part-number matching.The goal is to help teams avoid late-stage redesign by giving them better visibility into which alternatives are technically suitable, commercially available, and practical to qualify.<blockquote>“The most effective component strategies start with understanding the role each part plays in the system. For MLCC alternatives, the goal is not simply to find a replacement with matching specifications, but to identify solutions that maintain electrical performance, support qualification requirements, and provide flexibility throughout the product lifecycle.” </blockquote>— Zeb Zhan, Procurement Manager at Win Source<h2>From Capacitor Replacement to Capacitor Architecture</h2>MLCC alternatives are becoming more important because capacitor selection now sits closer to system-level design. Engineers are accounting for effective capacitance, impedance, bias behavior, thermal conditions, validation effort, lifecycle risk, and sourcing flexibility at the same time.The result is a more deliberate approach to capacitor architecture. MLCCs remain central to modern hardware, especially for dense, high-frequency designs. Alternative technologies expand the available design space by supporting bulk capacitance, rail stability, energy storage, ripple current, and long-term availability in positions where ceramics alone may create unnecessary constraints.For modern hardware teams, the practical question is whether the capacitor network gives the design enough electrical and supply-side resilience to move from prototype to production without unnecessary redesign. That makes capacitor strategy part of system design, and a useful lens for building hardware that can perform, scale, and remain supportable over time.

Polymer, tantalum, and other MLCC alternatives are giving engineers more design options to balance electrical performance, validation needs, lifecycle risk, and sourcing flexibility.

Beyond Shortages: How MLCC Alternatives Are Reshaping Modern Hardware Design

Narrowing 1,644 STM32 microcontrollers to one order code, filtered by design intent.

An STM32 microcontroller selection guide built as a working method: narrowing 1,644 parts down to one order code, organized by the constraint that actually drives your design rather than by part number.

Hardware Design

ORGANIZATIONS. SHAPING THE INDUSTRY.

Latest Posts

STM32 Nucleo Boards: How to Choose the Right One

STM32F103: Density, Packages, and the Blue Pill

AI Agents for Engineering

Beyond Shortages: How MLCC Alternatives Are Reshaping Modern Hardware Design

STM32 Selection Guide: Choosing the Right Part for Your Design