- Bricks & Bytes Bulletin
- Posts
- The Automation Paradox: Why 'Imperfect' AI Might Be the Only Honest Answer
The Automation Paradox: Why 'Imperfect' AI Might Be the Only Honest Answer
Owen Drury
August 20, 2025 • Estimated Reading Time: 15 minutes
Want to get your message in front of 1,971 highly engaged innovation leaders? Check out our sponsorship offers.
INDUSTRY INSIGHTS
The Automation Paradox: Why 'Imperfect' AI Might Be the Only Honest Answer
How one founder's 6-year journey from video games and high-performance computing to AI-powered electrical design reveals the hidden complexity of automation-first products
"Error-free design is not what we generate today."
This candid admission from an AEC tech CEO was a surprising one. Here's a company six years in the making, with a team building AI for electrical design, and their leader is telling us their system still requires human intervention.
But, after diving deep into Francesco Iorio's journey from video games development and computational research to automating electrical designs, we understand why this "imperfect" approach might be one of the most honest things anyone has said about design automation.
TL;DR: The Automation Paradox
AI in AEC doesn’t need to be perfect. It just needs to solve scarcity.
Perfect automation isn’t the goal. Solving real scarcity is.
80% solutions work if they make experts 5x more productive
Capacity expansion > efficiency gains = defensibility
Integration beats disruption in conservative industries
Deep domain expertise, not pure tech, drives adoption
Vision: democratize expertise so every building gets world-class design
The big idea? Imperfect AI that learns from human edits may be the only way to reach true automation — and transform entire industries.
The Fork in the Road
Every AEC tech company faces this moment: Do you build workflow acceleration tools or attempt end-to-end automation?
Most choose acceleration. It's the safer bet. Faster time-to-market, easier adoption, predictable revenue streams.
But some choose the harder path.
Design reviews in action. Source: Freepik
While competitors build point solutions that speed up existing workflows, certain teams spend years building AI that can route tens of miles of electrical conduit across complex buildings in minutes. Their approach? Getting "full results in a single shot, with minimal commands."
This decision ripples through everything: product architecture, development timelines, competitive positioning, and funding requirements.
Where others see workflow problems, they see automation opportunities
Here's the thing about that choice:
Autodesk's 2024 State of Design & Make report shows 44% of AECO professionals now cite improving productivity as their top AI use case. The industry is signaling readiness for more ambitious solutions.
But automation isn't just harder to build. It's a fundamentally different category of software that requires different thinking about market timing, competitive strategy, and capital requirements.
Selling Capacity Extension Vs Efficiency Gains
Six months into the automation journey, Augmenta.ai discovered something that transformed their entire approach.
The problem wasn't making electrical design faster. It was making expertise available at scale.
Think about it: Electrical contractors face a simple constraint. They need to complete projects on time with the people they have. But there simply aren't enough skilled professionals who do this job well.
You can't solve a talent shortage by hiring more people if those people don't exist.
But you can solve it by making existing experts 5x more productive.
When you're selling capacity expansion rather than efficiency improvement, you're not competing on cost. You're solving constraints that literally cannot be solved any other way. And that makes you defensible.
Automated routing and coordination of conduit systems. Source: Augmenta
The Counterintuitive Truth About "Perfect"
Here's where it gets interesting.
After six years of development, this system delivers designs that are roughly 80-85% complete. Users spend time finishing the remainder.
Your first instinct might be: "That's not automation, that's just assistance."
But here's the genius in the approach.
Every human intervention teaches the system. Every edit becomes training data. Every "incomplete" design reveals exactly where the AI needs improvement.
The team measures how much time people spend using their system relative to manual modeling, with the goal of driving that intervention time to zero.
The learning strategy only works under one condition: You must be solving a real scarcity problem.
If users desperately need the capability, they'll tolerate imperfection while you perfect the system. If they don't, they'll abandon imperfect automation for familiar workflows.
The generative design market validates this patient approach, growing from $298 million in 2024 to a projected $1.39 billion by 2034. Customers are willing to adopt imperfect automation that solves real problems.
The path to 100% automation isn't linear. It requires building learning systems, not just functional systems.
Playing Nice with the Ecosystem
Another paradoxical choice: integrate with existing platforms instead of building standalone solutions.
This creates fascinating dynamics. Automation companies operate within established ecosystems while potentially competing with those same platform providers' own automation initiatives.
Why choose integration over independence?
Because electrical contractors live in Revit. Fighting that reality adds years to adoption, regardless of technical superiority.
The trade-off is real:
Integration limits technical flexibility
But respects established user patterns
It's distribution strategy disguised as technical architecture
Acknowledges industry inertia while building for future possibilities
In conservative industries, distribution often matters more than differentiation.
You can have the most sophisticated technology in the world, but if it requires users to change fundamental workflows, adoption becomes an uphill battle that most startups can't sustain.
AI tools working inside familiar electrical design platforms. Source: Edraw.AI
The Expertise Bottleneck
What separates successful automation companies from the dozens of AI companies trying to crack construction?
Understanding both the technology AND the problem domain.
Technical backgrounds matter, but industry expertise creates the real differentiation. Teams with 20-30 years of electrical design experience understand pain points that pure technologists miss.
Domain expertise shows up in unexpected places:
Product decisions (starting with the most complex systems first)
Technical architecture (building for component physicality)
Go-to-market strategy (manufacturer-specific requirements from day one)
Customer conversations (hardly ever mentioning AI)
Successful automation companies make the technology secondary to solving real problems with deep industry knowledge.
The implication is sobering: you need either founding team domain expertise or the patience to develop it through years of customer development.
Pure technical talent isn't sufficient for complex automation. The rules governing electrical systems are "very quirky" and "very complicated." Some influence designs geometrically. Others affect designs in subtle ways that make computational solutions difficult.
The Infrastructure Vision
Here's what attracts sophisticated capital to automation companies: the promise of democratizing expertise.
Currently, only expensive projects can afford teams of engineers to optimize every system. Automation changes this equation.
Every building, regardless of budget, can have expertly designed systems.
This isn't just efficiency improvement. It's market expansion at infrastructure scale.
Think about historical precedent:
Throughout history, scarcity has been tackled by automation. Something only becomes plentiful and affordable when it reaches scale. Scale in most human endeavors has been accomplished through automation.
The vision extends beyond individual efficiency gains to industry transformation. Making "all the remaining 99.999% of buildings right across the planet just as performant" as the most expensive, expertly designed structures.
But it requires thinking in decades, not quarters.
The Reality of Building Hard Things
Six years feels like an eternity in startup time.
But building sophisticated automation requires accepting that complex software development is "inordinately expensive." The complexity isn't just technical; it's regulatory, practical, and cultural.
Electrical systems have written and unwritten rules. BIM teams often contend with field teams, and "at the end of the day, the field wins." Even proficient BIM designers sometimes lack the experience of master electricians.
This complexity creates defensibility.
The myth that big tech companies can easily dominate AEC spaces falls apart when you realize how hard it is for generalist technology companies to understand domain-specific requirements.
There's a component you can approach systematically and another you have to approach holistically. The learning from professional trades folks who have done this for decades isn't literally written anywhere.
Real automation that replaces human expertise rather than just accelerating human workflows requires different thinking about product development, market timing, and competitive strategy.
What This Means
The question isn't whether design automation will transform AEC.
It's whether companies are building for the reality of that transformation or the fantasy of it.
The reality requires patient capital and development cycles measured in years, not quarters. It demands deep domain expertise alongside technical capability, because pure engineering talent can't decode the unwritten rules that govern complex industries.
Success comes through learning systems that improve through imperfection, not products that launch as polished solutions.
The reality also means respecting industry inertia through thoughtful integration strategies rather than forcing users to abandon established workflows. And it requires thinking about market expansion rather than just efficiency improvement, because the biggest opportunities lie in making expert-level capabilities accessible to projects that could never afford them before.
The fantasy, by contrast, assumes that technical superiority alone can overcome distribution challenges. It believes that perfect automation is required for market success, when often the opposite is true. It imagines that standalone solutions can triumph over deeply embedded platforms, and that efficiency improvements automatically create venture-scale businesses.
The companies that understand this distinction are building the infrastructure for the next generation of construction technology. They're patient enough to solve hard problems properly, smart enough to work within existing systems, and ambitious enough to expand markets rather than just optimize workflows.
Watch the episode with Francesco Iorio here👇👇👇
What patterns do you see in your own automation challenges? Where do you think the biggest opportunities lie for end-to-end automation versus workflow acceleration? Hit reply and share your views.
WEEKLY MUSINGS
Power Race, Learning Gap, Cultural Friction
The AI boom has a huge cost
AI's a tool, but is it sticking?
The "us vs. them" mentality in construction
YOU MIGHT ALSO LIKE
Premium Insights
More Insights
Reports and Case Studies
Most Popular Episodes
Super Series
OUR SPONSORS
BuildVision — streamlining the construction supply chain with a unified platform for contractors, manufacturers, and stakeholders.