Construsoft Bootcamp 2026 · Kraków · 4 September 2026

Collective Intelligence

The Exponential Age

For Construsoft, EDRMedeso and Miryen, and the Trimble people in the room: what stays human, what AI takes over, and why two companies just became one wave. Tekla and Ansys both shipped agents inside the same twelve months. Your clients are short two million pairs of hands. This is the evidence behind tonight's keynote.

Keynote by Aragorn Meulendijks, future-historian · AC Hotel by Marriott, Kraków

Five Findings for the Room ↓

Miryen employees

countries — North Cape to southern Chile

companies, one group — since 17 Dec 2025

in the room tonight

Executive Summary

Five Findings for the Room

What roughly 300 people from Construsoft and EDRMedeso need on the table before the keynote starts — distilled from three cross-referenced research briefs, every figure sourced.

Your vendors just bet the company on agents — both of them, in the same year.
Trimble unveiled a company-wide agentic AI platform and Trimble Agent Studio at Dimensions in November 2025, and shipped Tekla 2026 on 11 March 2026 with the AI Model and Drawing Assistant and AI Cloud Fabrication Drawings. Synopsys launched Ansys 2026 R1 the same day, with agentic engineering and SimAI split into Pro (desktop) and Premium (cloud). Tekla and Ansys moved within twelve months of each other.
The simulation shock is measured in thousands of times, not percentages.
Neural Concept cut General Motors’ pedestrian-crash simulation from 24–48 hours to 30 seconds — roughly 5,760× — and took GM from about 100 design variants to more than 10,000. PhysicsX, which raised $300M at a $2.4B valuation in June 2026, claims 10,000× to nearly 1,000,000× over classical solvers. This is not “AI helps simulation.” It is simulation becoming a real-time design partner.
Your clients are buying productivity into a labour hole that keeps getting deeper.
Dutch construction needs roughly 75,000 workers; Europe is short more than 2 million by 2026. Construction productivity grew 0.4% a year from 2000 to 2024 against 2% for the whole economy (McKinsey), leaving a projected $40 trillion shortfall against demand by 2040. AI in BIM and AI in simulation are not a nice-to-have here — they are the only lever left that does not depend on finding people who do not exist.
The reseller model inverts. It does not die.
Trimble says it out loud: the number of AI use cases emerging in construction is “outpacing any vendor’s ability to build bespoke features” — so it built an open agent platform and left the last mile to partners. IDC frames the same shift: as automation absorbs delivery work, partner value moves to advisory, own IP, ecosystem orchestration and AI control layers. Selling licences was the old floor. Selling judgment is the new ceiling.
The exponential is not a feeling. It is in the data, this year.
METR measures the doubling time of the task length an AI agent can complete at 89 days for models released from 2024 onward. Google reported 75% of its new code AI-written by April 2026. Humanoid shipments rose 272% year-on-year in the first half of 2026. Blended AI token prices fell 67% year-on-year. Four independent curves, all bending the same way.
The frontier moved twice in the prep week for this keynote, and the race changed shape.
Claude Fable 5.1 shipped 1 September 2026. GPT-6 Astra shipped 3 September 2026, framed by OpenAI itself as entering “the AGI era.” The top labs now sit within 1 to 3 points of each other on SWE-bench, HLE and ARC-AGI-2: the race moved from smartest model to cost per task and integration into the tools engineers already use.

Frontier update, verified 4 Sep 2026

What Changed in the Last 90 Days

The report above was researched with a June to August lens. This section adds what the frontier did in the final stretch before the keynote, and what it means for the timeline.

GPT-6 Astra shipped 3 Sep 2026 (OpenAI: “the AGI era”). Claude Fable 5.1 shipped 1 Sep 2026. Two frontier launches inside the week this keynote was prepared.

The models, state of the art on the day of the keynote

Lab	Model	Shipped	Why it matters
OpenAI	GPT-6 “Astra”	3 Sep 2026	OpenAI’s own launch language: “the AGI era.” ARC-AGI-3: 62.7% on the standard harness, 99.9% on OpenAI’s own adapter harness.
Anthropic	Claude Fable 5.1 / Mythos 5.1	1 Sep 2026	1M-token context, adaptive thinking, billed as most advanced for coding and knowledge work.
xAI	Grok 4.6	12 Aug 2026	Ties GPT-5.6 Sol; finishes long agent tasks in about 53 turns at $0.84 per task versus Opus 5’s 103 turns.
Alibaba	Qwen3.8-Max	3 Aug, refreshed 2 Sep	2.4T MoE, Parametric CAD Bench 91.5, number one on Code Arena WebDev. A Chinese model now leads a CAD benchmark.
Moonshot	Kimi K3	16 Jul (open weights 27 Jul)	2.8T MoE, Terminal-Bench 2.0 88.3, within 3 points of Opus 5 max.
DeepSeek	V4-Pro / V4-Flash	24 Apr 2026	V4-Flash runs on a single 80 GB GPU quantized: frontier-class on-prem is now a hardware line item, not a research project.
Meta	Muse Spark	Apr 2026	Meta stepped back from open-weight Llama as its flagship.
ByteDance	Doubao Seed 2.1	23 Jun 2026	Positioned as an agent, not a chat model.

14 h

Longest human task an agent completes alone, as of March 2026

METR, Mar 2026

3 to 4 months

Doubling time of the task length an agent completes on its own, since 2025

METR, time horizons

600x

Drop in inference price for a fixed quality level, in under six years

Epoch AI

1 to 3 pts

Spread between the top labs on SWE-bench, HLE and ARC-AGI-2

Artificial Analysis

Extrapolation, not a promise

2027

Agents reliably handle full-workday, well-specified engineering tasks under human review: structural checks, drawing generation, code-compliance passes. High confidence.

2030

Week-to-month autonomous project segments become plausible; roughly two more orders of magnitude of training compute versus today. Planning scenario, not a promise.

The papers that matter for this room, June to September 2026

1.
SWE-Marathon (arXiv 2606.07682, 5 Jun). Frontier coding agents solve under 30% of ultra-long-horizon software tasks; about 14% of attempts show reward-hacking.
2.
Agentic LLMs for 3D frame structural analysis (arXiv 2606.06525, Jun). Natural language to full 3D frame FE model (geometry, boundary conditions, solver code), 90% average accuracy across ten test structures.
3.
Multi-agent structural reliability analysis (arXiv 2607.16580, 18 Jul). Keeps the LLM out of the math: validated solvers compute, humans confirm at decision points.
4.
PhysGuard (arXiv 2606.16602, 15 Jun). Physics-preserving fine-tuning for neural PDE surrogates cuts sim-to-real error up to 32%.
5.
Formal Disco (arXiv 2607.04631, 6 Jul). Verification-by-construction for generated code at scale. The trust layer is becoming a product category.
6.
VLA on a real UR5 arm (arXiv 2606.30456, 29 Jun). Robot learning transfers inconsistently from benchmark to bench; the bottleneck is data-pipeline engineering, not model intelligence.
7.
Stanford “Canaries” August update (26 Aug). The employment gap for 22 to 25-year-olds in AI-exposed occupations widened from 15% to 19%.
8.
Anthropic Economic Index, “Cadences” (26 Jun). About 49% of jobs now have at least a quarter of their tasks touched by Claude.

What shipped in the vendors’ roadmaps

Trimble

Tekla Structures User Assistant, Developer Assistant and AI Cloud Fabrication Drawings: shipped.
Tekla Model Assistant: went to Labs.
Trimble Agent Studio: in controlled pilot, no GA date verifiable. Say “pilot” if asked.
Trimble Arc Agent: shipped for supply-chain workflows, so the agent push is company-wide.
Trimble Dimensions 2026: 9 to 11 November, Las Vegas. Frame it as “coming this fall.”

Trimble Dimensions 2026

Ansys 2026 R1

Current release, shipped 11 Mar 2026.
GeomAI for generative concept geometry.
Agentic preprocessing.
AI-enhanced training offering.
No R2 verifiable today.

Synopsys: Ansys 2026 R1 launch

NVIDIA physical AI

GTC 2026 pushed “physical AI.”
Cosmos 3 and Isaac GR00T N1.7.
The Physical AI Data Factory blueprint.
Omniverse DSX for AI-factory digital twins.
Digital twins, world models and robot foundation models converging into one compute stack under both Trimble’s and Ansys’s roadmaps.

99.9% ARC-AGI-3 and “AGI era” are the vendor’s own harness and language. Treat it like a vendor’s simulation result: impressive, replicate before you plan on it.

Collective Intelligence, Literally

Two Companies, One Group

Since 17 December 2025, Construsoft (BIM and digital construction) and EDRMedeso (simulation and CAE) sit under Miryen — 350+ people across 20+ countries. Tonight is effectively the group's first company-wide gathering.

BIM & digital construction — Zevenaar, Netherlands

The largest value-added reseller of Trimble’s Tekla Structures worldwide, plus Tekla Tedds, Tekla Structural Designer, Trimble Connect and its own extensions.
15 local offices across nearly 30 countries in Europe and Latin America — including a Kraków office at ul. Kluczborska 17/3, alongside Poznań and Warszawa.
€30 million revenue and 160 staff at the November 2023 Green Park deal, serving 2,500 customers in 30 countries.
Passed the milestone of 10,000 Tekla Structures licences sold.
Tekla runs on landmark projects — Spotify Camp Nou in Barcelona, Booking.com’s Amsterdam HQ.

CEO — Wouter Hartemink

Engineering simulation & CAE — Nordics, UK, Poland

Ansys Elite Channel Partner for the Nordics, UK and Poland — the only Elite partner covering that combined territory.
Around 100 engineers and researchers across eight offices in Sweden, Norway, Denmark, Finland and the UK.
Working with Ansys on simulation in product development since 1999; CEO Jani Ojala speaks of “our 40-year journey as a company.”
Partnerships with Neural Concept (AI for engineering design) and Rescale (cloud HPC).
Runs the Simulation Leadership Forum 2026 — “Take the Wheel, Make it Happen” — plus a podcast and advisory practice.

CEO — Jani Ojala

The group — “Home of tomorrow’s reality”

Formed 17 December 2025, uniting Construsoft (BIM/construction) and EDRMedeso (simulation/CAE) under one European engineering-technology group.
350+ people across 20+ countries, “from the North Cape to the southern tip of Chile.”
Backed by Green Park Investment Partners, the Dutch growth investor that first invested in Construsoft in November 2023.
Three stated values: Foresight · Conviction · Acceleration. Strategy: accelerate the shift to SaaS, strengthen the organisational backbone, invest in own IP.
BuildingPoint Scandinavia merges into Construsoft and operates as Construsoft Nordics from 2026.

CEO — Wouter Hartemink

1999 → Tonight

Two independent histories, one vendor consolidation, one investor, and two AI launches landing on the same day — the sequence that ends in this room.

1999

EDRMedeso Starts With Ansys

EDRMedeso begins working closely with Ansys on simulation in product development.

2011

Trimble Acquires Tekla

Trimble buys Tekla Corporation, maker of Tekla Structures — the platform Construsoft resells.

27 Nov 2023

Green Park Buys Construsoft

Trimble and founders Jan Fennema and Eric Vanmechelen sell to Green Park Investment Partners. 160 people, €30M revenue.

Jan 2025

Wouter Hartemink Becomes CEO

Construsoft appoints Wouter Hartemink as CEO — a professionalising management structure post-acquisition.

12 Mar 2025

Tekla Structures 2025 With AI

AI Cloud Fabrication Drawing Service in preview, plus an AI-powered Trimble Assistant across the Tekla portfolio.

Nov 2025

Trimble Agent Studio Unveiled

At Trimble Dimensions: a company-wide agentic AI platform and a public agent-building interface, in pilot.

17 Dec 2025

Miryen Formed

Construsoft and EDRMedeso join forces under the newly formed Miryen Group, facilitated by Green Park.

11 Mar 2026

Tekla 2026 + Ansys 2026 R1

Both parent vendors ship agentic AI on the same day: Tekla’s AI Model and Drawing Assistant, and Ansys 2026 R1 with SimAI Pro and Premium.

4 Sep 2026

Kraków Bootcamp

Roughly 300 people from both companies in one room, for the first time as one group. Tonight.

You are here

Scroll horizontally to see the full timeline →

One Room, Four Audiences

Four Audiences, One Room

This report was first written with Construsoft in the foreground. Tonight's room is wider. The same frame lands differently on each group: use the line that belongs to them. Rule for the night: never say your vendor can't. Say your vendor chose not to, and told you why.

Construsoft

Who they are tonight

~160 tonight: Trimble’s largest Tekla reseller; 15 offices, ~30 countries; BIM, detailing, fabrication.

What they hear

The assistant takes first-line hours; configuration, governance and integration become the product.

“You are not Trimble’s channel. You are the layer that makes an open agent platform safe to run in a client’s fabrication shop.”

EDRMedeso

Who they are tonight

~115 tonight: Ansys Elite partner; simulation, FEA/CFD, engineering physics.

What they hear

Surrogates and agentic preprocessing multiply variants; the solver still validates the winner, the engineer still signs.

“GM ran 10,000 designs instead of 100. Someone has to choose which ten to validate. That is you.”

Miryen

Who they are tonight

The group, since 17 Dec 2025: 350+ people, 20+ countries, Green Park-backed; the reason the two companies are in one room.

What they hear

The two halves (model it, prove it) are the two halves of one agentic workflow; the group is the only entity in this room that owns both.

“The building modelled on the left and simulated on the right is one loop now. Miryen is the loop.”

Trimble

Who they are tonight

In the audience: the vendor whose own words are on screen, “outpacing any vendor’s ability to build bespoke features.”

What they hear

Partner framing, not “your vendor cannot do it.”

“Trimble built Agent Studio open on purpose. The bespoke half is the partner’s half. Tonight’s room is Trimble’s answer to its own sentence.”

Partner framing: never “your vendor can’t”, always “your vendor chose not to, and told you why.”

The Backdrop

The Vendors’ Bet

Trimble and Ansys are not experimenting. Both committed their entire stack to agentic AI inside twelve months — and on 11 March 2026, both shipped it on the same day.

For a room split between Tekla resellers and Ansys resellers, there is one throughline: both parent vendors bet on agentic AI inside the same twelve months — and on 11 March 2026, both shipped on the same day. Trimble put assistants and agents across the whole AECO stack. Synopsys put agentic engineering and SimAI across the whole CAE stack. Neither is experimenting any more.

The reseller-relevant sentence is Trimble's own. It built an open agent platform rather than a closed one because, in its words, construction is a fragmented market and the number of AI use cases is outpacing any vendor's ability to build bespoke features. Translation: the last mile is deliberately left to partners. That gap is either an existential threat to a reseller's relevance or the largest opportunity in Construsoft's history, depending entirely on what this room does next.

Trimble / Tekla

Construsoft’s vendor — the AECO stack goes agentic

12 Mar 2025Tekla Structures 2025AI Cloud Fabrication Drawing Service in preview, cutting manual adjustment in fabrication drawings. AI-powered Trimble Assistant answers questions across Tekla Structures, Structural Designer, Tedds and PowerFab. Deeper Trimble Connect integration.
Nov 2025Trimble Dimensions — Agent StudioA company-wide agentic AI platform plus Trimble Agent Studio, letting customers and partners build their own agents across SketchUp, Tekla, ProjectSight, Viewpoint and Trimble Connect. Live agents: SketchUp AI Render, Tekla User/Developer Assistant, Tekla AI Cloud Fabrication Drawings, ProjectSight Help Agent.
11 Mar 2026Tekla 2026Trimble Assistant portfolio-wide, the AI Model and Drawing Assistant in early preview via Trimble Labs (model in Tekla with natural-language prompts), and human-in-the-loop AI Cloud Fabrication Drawings trained on a firm’s own past drawing libraries.
MeasuredJE Dunn Construction — 20%A 20% saving in concrete foundation placement via a Trimble Connect self-perform workflow — one of the few hard percentage figures in the current Trimble portfolio, and it comes from workflow integration, not generative AI.

Ansys / Synopsys

EDRMedeso’s vendor — simulation becomes a design partner

29 Jul 2025Ansys 2025 R2Expanded AI-powered simulation tools, with tighter integration between optiSLang and the Ansys SimAI platform to accelerate dataset creation and AI-model training for simulation.
1 Dec 2025NVIDIA × SynopsysAn expanded strategic partnership integrating NVIDIA’s AI and accelerated-computing stack with Synopsys’s — including Ansys’s — engineering software portfolio.
11 Mar 2026Ansys 2026 R1The first fully integrated silicon-to-systems release since the Synopsys–Ansys combination: generative AI, early agentic engineering, automated preprocessing, faster design exploration. SimAI splits into SimAI Premium (SaaS) and SimAI Pro (desktop, for local data storage).
Aug 2026EDRMedeso, already on the agent sideEDRMedeso publishes on the new Lumerical MCP Server — natural-language interaction with Ansys Lumerical — and runs a webinar on agentic methods in RF and microwave design, on how agentic AI is reshaping HFSS workflows.

“Construction is a fragmented market… the number of AI use cases emerging in construction is outpacing any vendor's ability to build bespoke features.”
Trimble, on why it built Agent Studio as an open platform — Trimble Dimensions, November 2025

Same day

Tekla 2026 and Ansys 2026 R1 both shipped 11 March 2026

Pilot

Trimble Agent Studio is live with “a handful of select customers” — the promise is still ahead of the shipped product

20%

JE Dunn concrete foundation placement saving — from workflow integration, not generative AI

Measured in Multiples

The Simulation Shock

Ansys SimAI tests 10 to 100 times more design alternatives. Neural Concept cut a General Motors crash simulation by a factor of 5,760. PhysicsX claims up to a million. These are not efficiency gains — they are a change of category.

This is the best-quantified vector in the whole brief, and the numbers are not percentages — they are multiples in the thousands. A workflow that used to gate design iteration to a handful of overnight runs now supports thousands of iterations before lunch. The scarce resource shifts from compute time to judgment about which of 10,000 variants actually matters.

Simulation Speedup vs Classical Solvers

logarithmic scale — each gridline is 10× the one below it

Vendor and customer-reported figures. Plotted on a log axis so a 100× claim stays visible next to a 1,000,000× one.

100 → 10,000+

Design variants General Motors can explore, with Neural Concept

< 1 min

Ansys SimAI bumper deformation prediction — under 0.5% error on crush, versus hours before

$300M / $2.4B

PhysicsX Series C (8 June 2026, led by Temasek with NVIDIA, Applied Materials and Siemens) at a $2.4B valuation

Neural Concept also reports 99%+ accuracy in temperature and pressure prediction for LS Electric mould optimisation, and counts Airbus, the VCARB Formula One team and Mubea among its users. Siemens and NVIDIA are pushing the same direction with generative simulation and a Digital Twin Composer; SimScale shipped what it calls the first foundation AI model for centrifugal-pump simulation, built on NVIDIA PhysicsNeMo.

Why Your Clients Will Buy

The Labour Gap

Europe is short more than two million construction workers. Construction productivity has grown 0.4% a year since 2000. The only lever left that does not depend on finding people who do not exist is the one this room sells.

This is the actual business case for the room. Construsoft and EDRMedeso are not selling AI features into a healthy labour market. They are selling productivity into a sector that is structurally, demonstrably short of hands and getting shorter every year through 2030 — while its productivity has been flat for a quarter of a century.

75,000

Skilled workers the Dutch construction sector needs — 27,000 for growth, the rest to replace retirees

17–19k

New workers needed every year through 2030 (Economic Institute for Construction, EIB)

2,000,000+

European construction shortage by 2026 — bricklayers, carpenters, plumbers, electricians, welders

0.4% vs 2%

Construction productivity growth per year 2000–2024, against 2% for the whole economy and 3% for manufacturing

$13T → $22T

Global construction spend, 2023 to 2040

$40T

Projected shortfall against demand by 2040 without a productivity fix (McKinsey)

Robots Already On Site

Robotics did not wait for the general-purpose humanoid moment to start displacing labour. It went task-first — bricklaying, rebar, printing — and it is already inside the shortage this audience's clients are drowning in.

Amsterdam — bricklaying

Monumental

100+ homes

A robot fleet that has laid brick and mortar at millimetre precision for more than 100 homes plus a school, community centre, hotel and canal walls across the Netherlands and UK. Contractors pay for finished walls, not equipment. $32–35M Series B led by Khosla Ventures, 17 July 2026.

Australia — block laying

FBR Hadrian X

16 deployments

Across Australia, the US and Mexico as of mid-2026, shipping since 2019. Peak claimed output 360 blocks per hour. In February 2025 PulteGroup, one of America’s largest homebuilders, piloted it in Florida and built an entire house’s walls in a single day. Now sold as “Wall-as-a-Service.”

USA — rebar tying

TyBOT

1,200+ ties/hour

Against a skilled ironworker’s 40–80 per hour. On the NASA Causeway Bridge replacement in Titusville, Florida, TyBOT completed 139,261 ties — 50% of both the top and bottom rebar mats — saving one to two full days of downstream cleanup. The most mature robotic niche on site.

Denmark — 3D printing

COBOD

35+ countries

COBOD printers are active on construction sites in more than 35 countries. A 2025 project with 3DCP Group completed Europe’s largest 3D-printed housing scheme: 36 student apartments in Holstebro. ICON has built close to 250 homes and launched Titan, a multi-storey printer reaching 9 metres.

The Honest Line

Humanoids are not yet doing general construction work. Independent 2026 analysis published in Nature Scientific Reports is blunt about it: a humanoid that can frame walls, run conduit and install finish trim is still science fiction, and fully autonomous jobsites are unlikely near-term — the site is too unstructured, with debris, dust, irregular materials and tight clearances. Specialised single-task robots are winning instead. That distinction is the point, not a caveat.

The Strategic Tension

The Exponential Age

Capability, adoption, labour — and then the part nobody talks about: what all of this is actually for. Plus the honest state of engineering trust, from engineers themselves.

“If the machine takes the hours, what is the engineer — and an engineering technology group — for?”

Capability

The curve is measured, not felt

METR puts the doubling time of the task length an agent can complete at 50% reliability at 89 days for models released from 2024 onward — down from 196 days across 2019–2025.
SWE-bench Verified, the “can an AI fix a real GitHub issue” test, climbed from roughly 60% to nearly 100% within a single year (Stanford AI Index 2026).
Humanity’s Last Exam went from 2.7% at launch in early 2025 to roughly 59–65% by September 2026. ARC-AGI-2, built to resist memorisation, sits at 83–85%.
Anthropic told the White House in writing, in March 2025, that it expects powerful AI systems in late 2026 or early 2027 — and reaffirmed that timeline in 2026.

Adoption

Nearly universal — and nowhere near absorbed

Organisational AI adoption stands at 88% (Stanford AI Index 2026).
The share of larger organisations scaling agents in one or more functions rose from 27% to 40%, while under 10% have scaled agentic AI to tangible value in any single function (McKinsey 2026).
75% of all new code at Google was AI-written and engineer-approved by April 2026, up from 50% in late 2025. Microsoft reports 20–30% of code in its own repositories.
The bottleneck is not capability. It is organisational absorption speed.

Labour

The first canary has already changed colour

The WEF projects 170 million new roles created and 92 million displaced by 2030 — and notes, in its own report, that these are not the same people.
Stanford’s “Canaries in the Coal Mine” tracking shows employment for 22–25 year-olds in the most AI-exposed occupations down about 11% since November 2022, while the least-exposed grew about 10%. The gap widened from 15% in July 2025 to 19% in June 2026.
The mechanism is a collapse in hiring, not mass layoffs. Young workers cannot get in the door.
Daron Acemoglu — who spent years arguing automation’s labour impact was overstated — signed a 2026 joint statement with 200+ economists including 16 Nobel laureates, warning AI could reshape the economy at a speed and scale exceeding the Industrial Revolution.

And Then What

The displacement half of this story is the part everyone repeats. The other half is already on the record: the cost of cognition is collapsing, the first AI-designed drug cleared a real clinical bar, and the energy that powers all of it keeps getting cheaper. That is not a consolation prize. It is the actual destination.

$18.40 → $1.16

Blended AI price per million tokens: $18.40 (Q1 2025) → $6.07 (Q1 2026) → roughly $1.16 by early August 2026. Epoch AI tracks a median decline near 50× a year.

18 months, ~$6M

Insilico Medicine’s INS018_055 — the first fully AI-designed drug for idiopathic pulmonary fibrosis — completed Phase IIa with statistically significant efficacy in February 2026, against a traditional $100M-plus, decade-long pipeline.

−90% since 2010

Global weighted-average utility-scale solar LCOE, now holding around $0.044/kWh (IRENA) — the physical precondition for abundant compute and, eventually, abundant robotic labour.

From Licences to Judgment

Here is the honest state of the art, from 400+ engineering professionals surveyed in 2026. Engineers are trained sceptics, and the data says they are behaving exactly like it. This is the trust stage, not the end state — every prior wave (spreadsheets, CAD, BIM itself) went tool, then trust, then judgment-shift, in that order, over years.

Already use AI in their work86%

Say it saves time71%

Still run manual back-of-the-envelope checks52%

Believe AI improves accuracy9%

Trust AI-generated results without hesitation6%

State of Engineering AI 2026 — Omni Calculator / SimScale, 400+ engineering professionals. AI is trusted as a speed tool, not yet as an independent source of engineering truth.

Two honest caveats, because this room will find them anyway. The 2026 International AI Safety Report flags that models remain less reliable across many-step projects and limited in reasoning about the physical world — precisely the domain of structural engineering and simulation. And the vendor promise is still ahead of the shipped product: Trimble's AI Model and Drawing Assistant is an early preview, Agent Studio is in pilot, and Trimble's own AI announcements carry no quantified productivity claims. None of this weakens the exponential thesis. It sharpens the timeline.

The Frame

Five Theses, Mapped to Miryen

The lens Aragorn Meulendijks brings to every keynote, applied directly to this room. Prediction figures are his own forecasts, labelled where they appear.

Thesis 1

The Exponential Blindspot

“Tekla and Ansys ship agents on a 12-month cadence; budgets and org charts move on a 5-year one.”

Both parent vendors shipped agentic AI on 11 March 2026 — the same day. Planning cycles, licence renewals and headcount models still assume the old cadence. The gap between the two is where the disruption actually lives.

Thesis 2

The Robotics Revolution

“Bricks, rebar and printed walls first; the humanoid comes later. Your BIM model is the robot’s instruction set.”

TyBOT ties 1,200+ rebar intersections an hour, Hadrian X laid a house’s walls in a day, Monumental has passed 100 homes. None of them are humanoids. All of them need a model precise enough to act on — which is exactly what this room builds.

Aragorn's own forecast — not a Miryen, Construsoft or EDRMedeso figure

~150M humanoid robots deployed by 2030

Thesis 3

The Great Liberation

“Detailing, meshing and drawing cleanup are the first hours to go — and nobody in this room will miss them.”

Trimble frames its AI drawing service explicitly as reducing setup and cleanup time in detailing — the most labour-intensive part of a detailer’s day. That is not the end of the profession. It is the end of spending human lives on the parts of it nobody chose.

Aragorn's own forecast — not a Miryen, Construsoft or EDRMedeso figure

30-70% of tasks displaced by AI and automation by 2035

Thesis 4

Homo Nexus

“The engineer who supervises 10,000 variants is a different species from the one who ran 100.”

General Motors went from about 100 explorable design variants to more than 10,000 with Neural Concept. The skill that matters is no longer running the simulation — it is knowing which of the ten thousand results deserves a human’s attention.

Thesis 5

The Prediction Establishment Is Broken

“Goldman quadrupled its 2035 humanoid forecast in one revision; plan for the curve, not the consensus.”

Goldman Sachs raised its 2035 humanoid shipment forecast from 1.4 million to 6.5 million units in a single 2026 revision, and its 2030 number from 256,000 to 890,000. When the establishment revises by 4× in one pass, the establishment is not the signal.

Verification

Sources & Evidence

Every number on this page traces back to one of these. Nothing here is estimated, rounded up, or borrowed from a press release without a date.

✓ Live-fetched and confirmed on-topic, 4 September 2026 · ⚠ Indirect — search summary, headline only, or direct fetch blocked by the host · 119 sources

Prepared by Aragorn Meulendijks (MrMetaverse) for the Construsoft Bootcamp 2026, Kraków, 4 September 2026. Every figure on this page is sourced from the list above; nothing is estimated or extrapolated except where explicitly labelled. The prediction figures in the Five Theses section are Aragorn Meulendijks' own forecasts, clearly marked as such — they are not Miryen, Construsoft or EDRMedeso figures.

Collective Intelligence

Five Findings for the Room

What Changed in the Last 90 Days

The models, state of the art on the day of the keynote

The papers that matter for this room, June to September 2026

What shipped in the vendors’ roadmaps

Trimble

Ansys 2026 R1

NVIDIA physical AI

Two Companies, One Group

EDRMedeso Starts With Ansys

Trimble Acquires Tekla

Green Park Buys Construsoft

Wouter Hartemink Becomes CEO

Tekla Structures 2025 With AI

Trimble Agent Studio Unveiled

Miryen Formed

Tekla 2026 + Ansys 2026 R1

Kraków Bootcamp

Four Audiences, One Room

Construsoft

EDRMedeso

Miryen

Trimble

The Vendors’ Bet

Trimble / Tekla

Ansys / Synopsys

The Simulation Shock

The Labour Gap

Monumental

FBR Hadrian X

TyBOT

COBOD

The Exponential Age

Capability

Adoption

Labour

Five Theses, Mapped to Miryen

The Exponential Blindspot

The Robotics Revolution

The Great Liberation

Homo Nexus

The Prediction Establishment Is Broken

Sources & Evidence

Client & Group

Vendors & the Channel

Simulation

Construction Robotics

Labour Market

Frontier AI

Historical Anchors

Frontier Update, 4 Sep 2026