Director’s cut · Enterprise 3D · 2026
ATELIA, the long version
We wrote the user 3 times and renamed the idea 7 times. The idea itself barely moved.
The case study has the argument we ended up making. This is the week that produced it, including the strategy sprint we built and never used.
Role
Product strategy resident
Team
4: Melissa, Moe, Chloe, Krystal
Timeline
1 to 5 Oct 2026
Outcome
Pitched to Dassault leaders
8 min read · main cut 6 min

Iovine and Young Hall. 8 teams, 8 territories on the wall, 5 days.
20
dated signals, 5 from each of us
4
worlds on the quadrant, 1 each
7
names for the same idea, ending in ATELIA
9
recurring concerns from 4 rounds of peer review
A The territory
“The Room That Designs Together”
8 teams, 8 territories. Ours arrived as 3 questions, and we were not allowed to narrow it for the first 2 days.
What does the virtual twin become when it’s a shared workspace people think inside together, rather than a deliverable handed from one person to the next? How do decisions get made when everyone is in the model at once? And what is lost or gained when design becomes a live, collective conversation?
From the brief
B Signal harvest
A signal has a date, or it is not a signal
5 signals each, 4 of us, one research lane apiece. The brief set the bar and told us what would happen to the ones that missed it.
The rule
A signal is something that has already happened and carries a date: a shipped product, a regulatory phase-in, a funding round, a patent grant, a price curve.
Not a signal
✗ “Companies are investing in AI.” No date, no event.
A signal
✓ “The EU AI Act’s high-risk obligations phase in on a published schedule.”
My lane was the maker and prosumer economy
5 signals, every one dated and cited to a primary source rather than a press write-up.
01
1 Jul 2025 · Figma S-1. Non-designers made up two thirds of Figma’s 13 million monthly users. The busiest shared design workspace is already mostly people who aren’t designers.
02
Feb 2026 · Roblox annual report. Creators earned $1,503.1M in 2025, up from $922.8M. Real-time co-creation already works outside professional tools, at $1.5B a year.
03
14 Jul 2026 · PTC Onshape Labs. AI agents and a text-to-CAD server. Onshape’s free plan includes real-time collaboration, as long as designs stay public. A competitor gives makers multiplayer CAD in exchange for openness.
04
29 May 2026 · Creality lists in Hong Kong. About $177M raised, and Creality Cloud has 6.2 million users. The prosumer loop is now a publicly audited industry running entirely outside professional CAD.
05
24 Mar 2026 · Washington ESHB 2320. From 30 Jun 2027, distributing gun CAD files to unlicensed people is an infraction. The design file itself is now regulated, so whoever hosts a shared model takes on liability that differs by state.

The signal board, 1 column per person. Mine is the green run on the left.
C Forces and bets
A force has a verb in it
The rule again: a force is a direction of travel inferred from clustered signals, written as a sentence with a verb in it. Not a topic. Not a prediction nobody could disprove. 20 signals in, 4 forces out.
Force 1
Consolidated end to end service, optimised with automation and available on demand, becomes the norm.
Force 2
Intuitive low cost creation platforms let non-professional designers and creators in.
Force 3
Human teams and agentic twins co-work on shared models in real time, connecting design decisions to engineering analysis and simulation at the same moment.
Force 4
Databases and cloud services are high value assets in high demand, and they need better protection.
The one I should have plotted instead
Also in my 5, and not the one I picked: new AI-native creation tools let people learn and build, but paywalls will stop underserved communities from upskilling and joining the workforce. It is the only uncertainty of mine that survived into the final argument, because the whole pricing model is an answer to it. Spend the tokens you already have, nothing new to buy.
D The quadrant
4 worlds, 1 each, and we did not pick mine
2 axes: international collaboration against international competition, and whether AI or people are responsible for the higher-level thinking. 1 world written per person. Each had to say what is true here and who wins or gets locked out. It also needed a day in the life and a bar another team could judge a concept against.
Collaboration + AI-led
The Global Brain
Countries and companies share AI agents and knowledge globally, solving faster while becoming dependent on a network no single participant fully understands.
Collaboration + human-led · we built here
Intellectual Harmony
Technology takes the routine work and experts collaborate across borders, but people keep the judgment. The only one of the 4 where a knowledge exchange is the obvious product rather than a liability.
Competition + AI-led
The Algorithmic Arms Race
Nations treat design intelligence as a strategic asset. Machines decide, and people mainly execute decisions already made, then refine and justify them.
Competition + human-led · my world
Adopt Me
A national AI race. Every leader wants to run the most advanced model in the world, and the day in the life follows a founder racing to get his adopted abroad before a rival nation overtakes him.

The quadrant. Our world sits top right: expertise crosses borders, people keep the judgment.
E Who it is for
We wrote the user 3 times
The persona changed more than the concept did, and each change narrowed the problem.
First
Jordan, 18
Engineer for a high school robotics club in Los Angeles. “I want more opportunities to meet robotics experts and build my start-up.” She does not know how to navigate production, suppliers or manufacturers.
Second
Ty, 45
Product development engineer with an international team, flying constantly. “I want consistent information at all times to guide decision-making.” Time zones and language get in the way, and clients hold different versions of the same proposal.
Third, and the one that stuck
Donald, 33
Electrical engineer at a small hardware startup in Los Angeles. “I need expert-level guidance without hiring a specialist for every design decision.”
Jordan and Ty are both coordination problems: too many people, not enough shared context. Donald is a capability problem: the knowledge exists and he cannot reach it. Swapping the persona is what turned the concept from a better meeting room into a marketplace.
F How it narrowed
7 names for the same idea
The board records the whole chain. Each step is smaller than the one before it.
01
One-Stop Build Network. Everything needed to take hardware from idea to market, with AI recommending the production path.
02
Shared AI, Protected IP. Industry models that improve on contributed data without exposing anyone’s IP.
03
Dassault as the one-stop shop for hardware. The 2 merged. Our most ambitious version, and the one peers pushed hardest on.
04
The Dassault Build Network. The same, plus a named human signing every compliance decision.
05
Dassault Production Path Advisor. Narrowed to one job: recommending where and how to make a product.
06
The Process Exchange. Narrowed again, to licensing the know-how rather than brokering the suppliers.
07
The AI-Powered Skills Path, which became ATELIA. The same exchange, with each licensed process running as a skill inside 3DEXPERIENCE.
The note where it turned
The pivot is a single unformatted sticky, typed fast and never cleaned up: our solution companies to contribute and share knowledge · data protection, prevent leakage · already have 3DEXPERIENCE AI Skills · based on the companies skills · incentive · selection of who receives data, which data · trusted partners · skill-use basis (royalties, partial ownerships) · Our proposal is a skills marketplace for Dassault’s consumers to selectively provide their unique skills
G The cut that hurt
The strategy sprint we built and did not use
Anchored in Intellectual Harmony, then flexed across the other 3 worlds. 7 strategy cards, a pilot, a measurement plan and a 4-beat pitch. The section on the board is titled, in our own words, “Strategy: The Co-Creation Guild, didn’t really use”.
Companies don’t refuse to collaborate. They refuse to collaborate for free, at risk, and at extra cost. Flip all 3 and connection becomes the default.
The one line from the whole week I would still defend.
3 layers
Safe, so sharing cannot hurt you. Pays, so sharing earns you money. Flows, so sharing saves you time and cost. 7 cards hung off those 3, including “Share the Behavior, Not the Blueprint” and “Contribution Ledger: Credit = Income”.
Why we dropped it
It was a strategy for an industry, and the brief asked for a product. ATELIA’s 2-way seal is the safe layer and the token split is the pays layer, so most of it survived in compressed form. I built it with AI assistance, and the card on the board says so.
H Peer review
4 rotations, 9 recurring concerns
4 rotations of 5 minutes each with Groups 1 to 4 on the morning of 4 October, recorded and then synthesised. I did the synthesis, with AI help on the transcripts, and the board carries that note.
Why would anyone share?
Each level of sharing has its own payoff: orders, licensing revenue, tokens, discounts, better matches.
What is actually new?
Lead with the business model. Our scan found no CAD or product-data vendor that rewards contributions.
Privacy and the big data pool
No pool. Data goes to chosen partners by default and never raw to competitors, and prices are never pooled.
Can a process even be owned?
Package it as a licensable file: machining setups, simulation templates, inspection plans.
How do you measure contributions?
Reward use, not volume, and cap any one company’s share of the rewards.
Bias from who opts in
Track which industries and regions are covered, and fill the gaps through fab labs.
Who approves what comes in?
Each contributor sets the permissions on a data passport, and Dassault approves every listing.
What would the data look like?
A standard capability profile: processes, materials, tolerances, certifications, capacity, lead time, region.
Scope
Organise information and matching. Do not try to replace the conversation between people.
I Mentor notes
Their notes as taken, not tidied up
3 minutes to pitch, 9 for feedback, 8 for co-ideation. These are the notes close to as taken, not tidied into conclusions.
Abhishek Bali · Director, Mainstream Innovation
Study FabLabs and maker spaces, Dassault works closely with them. Anyone with 3D fabrication tools can start one. They are public workshops where people come together: how can they provide tools for creation. Look at Fab Foundation, Fab Lab Rwanda, Artisans Asylum. Each lab funds itself. Dassault finds success with people who know what they want and ask for it. Startups currently work in a low-key, individualistic way. Experts who stay in one lane help startups most. Many of them come out of a PhD thesis.
Suchit Jain · VP, Strategy and Business Development
Co-creation where the boundaries of companies do not exist. Asynchronous against synchronous collaboration. Supply chain and IP: people do not want to share. Surrogate modelling, so you teach the system your own data. Frontier models are fixed and public; with NVIDIA’s neutron models you can build your own natively inside the company. Industry world models.
J The standard
If it needs you in the room, it is not finished
1 sentence appeared on 3 of our 4 world cards and again at the end of the strategy board. It did more work than any framework we used.
Another team must be able to pick up your card on Sunday and judge someone else’s concept against it without you in the room. If it needs you to explain it, it is not finished.
On 3 of our 4 world cards

The cohort’s worlds on the studio wall. We tested ATELIA against every one of them.
K What I’d do differently
2 days lost to the wrong uncertainty
3 process notes, not product ones. The product gaps are in the case study.
I plotted the wrong uncertainty
My paywall uncertainty is the one the final business model answers. I plotted the robotics one because it felt bigger, and lost 2 days of framing because of it.
We converted forces too fast
The regulation cluster never became a sentence, and certification turned out to be central. The conversion pass was a timed exercise and we treated the timer as the goal.
The Guild should have been cut a day earlier
It was good work aimed at the wrong deliverable. The signal was there the whole time: the brief asked what Dassault could build, and a guild is not something you build.
Want the short version?
The case study
The 6-minute cut has the argument we made and the prototype we pitched to Dassault.
Keep reading · case study 2 of 5
Founder project · Consumer goods · 2025–26
MOKIO
Most bags tell you the country. MOKIO tells you the village.
© 2026 Moe Htet
Los Angeles