Summary of "Joe Justice ABB Keynote 2024 09 11"
High-level summary
- Joe Justice (founder of Wikispeed, former Tesla agile lead) presented an engineering‑led model for “agile hardware”: short, continuous hardware iterations enabled by modular design, shop‑floor work, heavy automation, and AI‑assisted group work.
- Core claim:
Replace traditional project plans/schedules with a real‑time product/dashboard for each modular component (the “Justice board”), where leadership sets KPIs per module and engineers self‑select daily work to continuously improve those KPIs.
Key technological concepts and practices
Justice board (dashboard)
- One row per module (modules = independently developable pieces of a product).
- Leadership defines modules and KPIs (cost, weight, power, cycle time, reliability); engineers choose a row and improve it day‑by‑day.
- Real‑time metrics reported by sensors and factory systems, visible company‑wide (large monitors + mobile app).
Modularity (Justice’s law)
- Keep products split into about <10 independent modules so teams can work in any order without cross‑team scheduling.
- Modular boundaries can change as teams discover better splits (inspect & adapt).
Daily / rapid hardware iterations
- Aim to incrementally improve parts daily (grams lighter, pennies cheaper, milliwatts less) instead of multi‑year release cadences.
- Start with current iteration length and shorten via automation, reorganization, and modularization. Examples: SpaceX/Tesla moved some flows from years to hours/minutes.
Mob AI (AI‑assisted group work)
- Small teams (3–5 people) work together on the factory floor or remotely and use AI continuously to ideate, design, and automate production steps.
- Practical AI training method: whenever an external AI gives a useful answer, record three items—question, context, response—to iteratively train an internal AI.
- Internal AI is trained on CAD + production steps + daily variations; it can generate designs, production steps, and anticipate certification questions.
Factory‑centered prototyping and automation
- Engineers work directly in the factory: manually prototype, then reprogram production robots to trial changes on ~10% of lines. One cell can become a daily prototype channel.
- CAD and production‑step models are created after manual prototyping to train AI and automate production.
Embedded and continuous testing
- Parts and products self‑test during assembly (embedded tests in “factory mode”), rather than relying on occasional separate testers.
- Test‑every‑unit approach enables safe frequent changes, rapid field fault detection, and supports frequent small certification submissions.
Certification and regulatory strategy
- Group changes by whether they require regulatory re‑certification; many incremental changes do not trigger full re‑certification.
- Submit small, frequent approval requests with tests and analytics instead of infrequent large submissions; feed certification outcomes into internal AI to optimize future submissions.
Internal tooling and vertical integration
- Value rises when internal AI can access company CAD, CAE (magnetics, CFD), production‑step data, and in‑house drawing/automation tools.
- Tesla cited as an example of building many internal tools to support this model.
Scaling reuse and economies of scale
- Reuse modules across product lines (e.g., seats, motors, heat pumps, inverters) because the Justice board makes KPIs visible across teams.
- Large installed volumes accelerate learning and AI training (billions of operational hours provide rich data).
Case studies / examples
- Wikispeed: early 1‑week iteration car development that inspired agile hardware ideas.
- Tesla:
- Justice‑style board and daily KPI improvements across modules.
- Heat pump: evolved to a tiny, highly efficient design via daily factory prototyping, robot reprogramming, and AI training.
- Reuse: shared seats, motors, heat pumps, and inverters across models; real‑time analytics; Model 3 costs reportedly down ~30% over four years.
- Internal AI trained on CAD + production steps suggesting designs and production sequences.
- SpaceX:
- Hundreds/thousands of rapid hardware changes and high automation (e.g., integrated X‑ray in welding robots).
- Iteration time reductions: multi‑year → months/days → sub‑hour loops in specific processes.
- Neuralink, The Boring Company: adapting rapid‑iteration/automation patterns to medical devices and construction.
- Toyota, Honda, Sony, Hitachi: piloting similar approaches; Toyota tested few‑module starts to cut development/plant investment via modularity.
Organizational and people implications
- Leadership: decide product modularization and KPIs; replace PM schedules with the Justice board and empower engineers to self‑organize.
- Culture: flattened hierarchy where engineers work alongside production; “everyone is a worker” philosophy.
- HR / talent:
- Mission‑driven hiring advantage (high applicant numbers at Tesla/SpaceX).
- Mob AI broadens accessibility across ages and skill levels; senior staff can mentor AI and juniors while remaining productive.
- HR must operate as a strong service function to support the model.
- Work practices:
- Prefer small in‑person mobs within ~3 meters for close coordination; remote mobs are viable and reduce commute (in‑person slightly faster).
- Many traditional roles (project managers, architects, heavy schedulers) are reduced or redefined.
Practical roadmap (compact)
- Leadership: replace project Gantt schedules with a Justice board listing products → modules → KPI per module.
- Define modules (allow later adjustments); limit number of modules per product.
- Make modules measurable with real‑time factory sensors (cost, weight, power, cycle time, reliability).
- Move engineers into module rows; support 3–5 person mobs working daily in/near production.
- Start small: pilot with a few modules (e.g., Toyota began with three).
- Train internal AI continuously from work artifacts (question/context/response) and CAD/production‑step data.
- Embed tests in each unit and adopt test‑every‑unit to enable safe frequent changes.
- Automate repetitive tests and production steps; use robot reprogramming to trial daily prototypes.
- Classify changes by certification need; submit frequent small approvals and train AI on approvals.
- Repeat, scale module reuse across products, and iterate to shorter cycles.
Guides, workshops, and resources mentioned
- Joe’s 12‑step workshop: two days (12 hours total) covering practical steps for transformation.
- Joe DX class: online, two 6‑hour days (deeper, paid offering).
- Practical training on mob AI and the Justice board method delivered globally.
Limitations, cautions, and prerequisites
- Leadership commitment is required (a clear decision to replace project scheduling with a dashboard).
- Significant investment needed in sensors, internal AI, tooling, and factory integration.
- Cultural change: engineers must be willing/able to work in factory environments; safety and certification processes need rethinking.
- Not every company can implement all aspects immediately; conservative, incremental adoption (as Toyota demonstrates) is a viable path.
Main speakers / sources
- Joe Justice — keynote presenter; founder of Wikispeed; former Tesla agile practice founder; consultant to Toyota, Honda, and others.
- Event host / organizers: ABB, with moderation/introductions by ABB staff (including “Charles” and teams in Sweden).
- Referenced organizations and influences: Tesla, SpaceX, Neuralink, The Boring Company, Toyota, Honda, Mercedes‑Benz, BMW, and mentions of Elon Musk and Akio Toyoda.
Category
Technology
Share this summary
Is the summary off?
If you think the summary is inaccurate, you can reprocess it with the latest model.
Preparing reprocess...