Synthesis — what the framework predicts about AEC software in 2026

The pattern across the war games

The framework's analytical contribution — the reason a 10-moat lens matters in AEC software in 2026 — is that the AI-era moats reshape which attackers survive consolidation, which get acquired, and which die. The verdicts cluster: attackers anchored on Counter-Positioning plus a decision-bearing Data Flywheel stay independent (Higharc, Trunk Tools, GleanCo-AEC, HarveyCo-AEC); attackers whose moat is generation-side capability without an evaluator surface get acquired (Snaptrude, Buildots+OpenSpace); attackers at the API/integration perimeter face the cleanest acquired-or-independent fork because the incumbent's defense option is binary (Speckle, PlaidCo-AEC); and the multiplayer-network play (FigmaCo-AEC) lands acquired in the base case because cross-firm AEC coordination requires owner mandate and owners do not mandate tools. The shape is not random and it is not pre-decided by capital — it is determined by which combination of the ten moats the attacker actually has, against which classical incumbent defense is binding.

The survival prediction table

The asterisks (*) on Speckle and GleanCo-AEC mark cells whose verdict is most sensitive to the four paper-wide forward-tracking bets covered below. Higharc and Speckle profiles are locked from earlier war-gaming. Fictional attackers (GleanCo-AEC, HarveyCo-AEC, DecagonCo-AEC, FigmaCo-AEC, PlaidCo-AEC) are framework-generated company designs, not real 2026 companies; their valuations are conditional on the attacker shipping in the lane the framework predicts is open.

Pricing-as-counter-positioning — the most under-applied moat-formation move in AEC software 2026

The AEC-software incumbent map is dominated by one pricing model: per-seat. Autodesk built it. AutoCAD, Revit, Civil 3D, Inventor, the entire Forma stack — all priced per-seat. The seat license is the largest single revenue mechanism in AEC software and the deepest single source of cannibalization risk: any product Autodesk ships that disintermediates the seat (per-outcome, per-project, per-customer-volume) breaks the revenue line that funds Autodesk's ability to ship anything at all.

BricsCAD has run this attack since the mid-2000s. Same product surface as AutoCAD, perpetual-license pricing instead of subscription, ~70% price arbitrage. Bricsys stayed independent for two decades and was acquired by Hexagon in 2018 for $400M+ — not by Autodesk. The acquirer pattern is the proof: Autodesk could not buy Bricsys without conceding that perpetual licensing was a viable wedge in its own market. Twenty years of independence on a single pricing-arbitrage thesis.

The AI-era extensions sharpen the lineage with four pricing archetypes contending in 2026:

• Per-outcome. Joist AI (per-proposal-produced); Higharc (per-home-sold); HarveyCo-AEC (per-evaluation, share-of-savings on E&O premium reduction). The vendor takes outcome risk; the customer pays at the level of value created, not at the level of usage.
• Per-customer-size. Procore (annual-construction-volume tiers). Travels into AEC because the underlying customer-size proxy — jobs-on-the-books, GMV processed, homes-sold — is observable and contract-able.
• Consumption-based. Per-token, per-API-call, per-document-processed. DecagonCo-AEC's mid-market wedge sits here. Counter-positions Procore's upmarket-volume model directly because consumption pricing serves customers Procore prices out.
• Per-seat. The classical Autodesk model. Every dollar Autodesk spends on AI capability lands on top of the seat license, accelerating the cannibalization Autodesk already cannot absorb. The classical defense is a melting-ice-cube bet: the longer the per-seat model holds the dollar-weighted middle, the more value flows around it through per-outcome and consumption-priced wedges.

The argument compresses to this: if the incumbent's pricing model is structurally locked to seat-licensing, every AI-era pricing innovation is a counter-positioning attack the incumbent cannot follow without breaking its own revenue line. Counter-Positioning chapter 13 named this as a moat that strengthens with incumbent AI investment. The pricing-arbitrage lineage shows it has worked for 20 years already; the AI-era extensions just give it sharper teeth.

What an AEC-native attacker should look like in 2026

Given the framework, the AEC battlefield, and the horizontal archetype DNA, here are three specific design archetypes that the framework predicts should exist and largely don't yet. These are not enumerations of all possibilities — they are the three the framework most loudly recommends, and they are different from the existing real attackers.

Archetype 1 — The AEC Evaluator-Power play, anchored on insurer underwriting

Layer to attack: Priya's layer-3 (accountability/stamping) plus layer-2 (engineering judgment), in a high-stakes professional-services slice where the consequence of being wrong is severe and the reviewer corpus is buildable cross-firm. Best wedge: jurisdiction-specific code-compliance review (egress, energy, accessibility, structural code) for mid-to-large commercial buildings.

Pricing model: share-of-savings on E&O premium reduction, in partnership with one named insurer (Travelers, Berkley, Beazley) underwriting a narrow product. Per-evaluation pricing as the secondary tier for non-insured customers. The pricing model is the moat — per-seat would be wrong, per-outcome ties the vendor to the customer's actual risk reduction.

Moat composition: Foundational Evaluator Power (layers 2 and 3); foundational Data Flywheel on the cross-jurisdiction code-interpretation corpus. Supporting Switching Costs once the AHJ workflow runs through it; supporting Brand once the insurer partnership is announced.

Wedge customer: mid-market AE firms (50–500 employees) doing commercial work in 2–5 jurisdictions; one progressive AHJ pilot program; one named insurer underwriting partnership.

Bull-case 3-year valuation: $2–5B if the insurer underwriting hits in 2027 and the AHJ pilot expands to 3+ jurisdictions by 2028.

Single biggest risk: insurer underwriting does not happen in the 3-year window. Without it, the company is a per-evaluation tool, not an evaluator-power business.

Archetype 2 — The federated agent network for the project lifecycle, on consumption pricing

Layer to attack: the project-lifecycle workflow at mid-market GCs (50–500 employees) priced out of full Procore implementations. Specialist nodes: submittals, RFI generation/response, change-order analysis, daily reports, payment-application review, lien-waiver tracking. Federated through MCP-style tool calls, deep at one job each.

Pricing model: per-task or per-token consumption pricing. Counter-positions Procore's annual-construction-volume model directly. The mid-market wedge is real and structurally protected by Procore's cannibalization bind — Procore cannot reprice for mid-market without torching the upmarket land-and-expand engine that funds it.

Moat composition: Foundational Agentic Workflow Lock-in (tool-graph + calibration depth across the federated nodes); foundational Counter-Positioning on consumption pricing. Supporting Data Flywheel on per-task corpus accumulating cross-customer.

Wedge customer: mid-market GCs running document-heavy operations on email/Slack/SharePoint/Excel without an integrated platform. Estimated TAM: 4,000–6,000 GCs in the US in this segment; ~$800M–$1.5B annual software-spend headroom.

Bull-case 3-year valuation: $1.5–3B at $200–500M revenue if the company stays disciplined in mid-market and accumulates calibration depth before Procore reprices.

Single biggest risk: Procore launches a consumption-tier mid-market offering before the federated network anchors. The cannibalization bind is real but not absolute; if Procore decides the mid-market is worth a partial price-model fracture, the wedge closes.

Archetype 3 — The counter-positioning play at the API perimeter, on per-project pricing

Layer to attack: the project-graph layer Autodesk's APS gates. Speckle is the open-source incumbent here but is structurally mid-acquisition or pre-acquired; the lane is open for a commercial-first attacker that ships a per-project-priced project-graph hosted layer.

Pricing model: per-project pricing. The customer pays a flat fee per project committed to the platform as system of record, not per-seat and not per-API-call. The cannibalization bind on Autodesk's APS is sharper here than on Speckle: Autodesk could plausibly compete with an open-source offering by tiering APS, but it cannot compete with per-project pricing without cannibalizing the seat license that funds the entire AEC segment. This is the cleanest post-Speckle play.

Moat composition: Foundational Counter-Positioning on the per-project pricing-arbitrage lineage (Bricsys precedent). Foundational Data Flywheel on the project-graph itself — every project committed becomes a source of decision-bearing data and a logo other tools build natively against. Supporting Network Economies via the federated-tool ecosystem (chapter 09 reframe).

Wedge customer: progressive AE firms and tech-leaning GCs (Suffolk, DPR, NBBJ, CRTKL) running 5–50 active projects at any time. Per-project pricing is contract-able directly against project-budgeted software spend, which is a different finance pocket than per-seat licensing.

Bull-case 3-year valuation: $2–4B if the company anchors 50+ enterprise customers and 500+ projects committed by 2028, before Autodesk decides to ship a free APS tier.

Single biggest risk: Speckle wins the AI-integration mindshare in 2026–2027 and the post-Speckle commercial lane never opens. This archetype is conditional on Speckle being acquired or stalling out before commercial-first attack hits a critical mass.

The four asterisks as forward-tracking bets

The four paper-wide asterisks are load-bearing matchups whose H2 verdict could not be cleanly called from current evidence. They are forward-tracking bets — positions whose resolution will sharpen the framework's predictions over the next 12–36 months. Each is named below with the specific signal that updates it.

Asterisk #1 — Glean vs. Microsoft Copilot at H2

The cleanest test of Data Flywheel beats Speed once spinning (chapter 13's loud claim 1). Microsoft has the scale, distribution, and capital; Glean has the 12–18-month enterprise knowledge-graph maturation curve. Load-bearing because: if the flywheel beats scale, the AEC analog (GleanCo-AEC) is in the lane the framework predicts; if scale beats the flywheel, every Data-Flywheel-anchored attacker in the cast (including Higharc and HarveyCo-AEC) faces a tougher H2 than predicted. Updates on: 2027 enterprise-search market-share data showing Glean's growth rate inside Microsoft tenants vs. Copilot's capture rate of Glean's ICP. If Glean's daily-default share at top-100 enterprise accounts is rising in 2027, the flywheel is winning; if Microsoft is converting Glean accounts to Copilot, scale is winning.

Asterisk #2 — NetSuite vs. AI-native ERP greenfield at H2

The canonical example of Switching Costs amplified by Agentic Workflow Lock-in (chapter 13's loud claim 2 — the multiplicative-composition claim). Embedded SaaS plus agentic calibration on top is structurally more defensible than either alone. Load-bearing because: if NetSuite + AI agents holds against AI-native ERP greenfield, the Procore-vs-DecagonCo-AEC matchup tilts toward Procore + agents winning, and the mid-market consumption-priced wedge narrows. If AI-native ERP wins greenfield against NetSuite, the federated-agent-network archetype runs hotter than predicted. Updates on: 2027–2028 net-new ERP wins at companies with $50M–$500M revenue. NetSuite + AI capture rate vs. AI-native (Rillet, Campfire, Lava) capture rate is the proxy.

Asterisk #3 — Speckle vs. Autodesk reframed (API/data-gravity, mindshare timing window)

Counter-Positioning under AI pressure, reframed around the API perimeter. Autodesk wins H2 by shipping a credibly open project-graph at the cost of Revit-license margin, or by acquiring Speckle to neutralize the mindshare threat. Load-bearing because: the Speckle outcome determines the post-Speckle lane (Archetype 3 above). If Speckle is acquired by 2027, the per-project commercial-first lane opens; if Speckle goes independent at $1–2B, the lane is closed because Speckle owns the open mindshare and the commercial-first play is structurally flanked. Updates on: Autodesk APS pricing changes at AU 2026 or AU 2027; any Autodesk-Speckle acquisition rumor or filing; Speckle's commercial-tier ARR if it discloses.

Asterisk #4 — Suffolk vs. owner-led IPD at H2

The structural test of whether owner-led IPD with AI-native preconstruction tools displaces tech-leaning-GC scale-plus-process moats on hyperscaler datacenter portfolios. Load-bearing because: if hyperscaler owners run their own IPD, Turner's scale-on-the-tailwind thesis softens at H2 and FigmaCo-AEC's bull-case-INDEPENDENT path opens (a repeat-builder owner mandating a multiplayer coordination platform is the same signal). If hyperscalers continue contracting through tier-1 GCs, the owner-side coordination layer stays fragmented and FigmaCo-AEC stays in the acquired base case. Updates on: 2027–2028 announcements of hyperscaler internal-construction-management capacity expansion (Microsoft's Modular Construction Group, Meta's internal data-center program, Amazon's AWS Infrastructure Construction). Headcount + scope expansion is the proxy.

Implications for an AEC operator or investor in 2026

Five specific bets the framework recommends, ranked by conviction.

1. Build or back the AEC Evaluator-Power play with insurer underwriting. Archetype 1 above. The lane is empty in 2026 and the framework predicts $2–5B 3-year value if the insurer hits. The work in 2026 is two-fold: (a) anchor a partnership conversation with one of Travelers / Berkley / Beazley around AI-assisted code-compliance review; (b) build the cross-jurisdiction code-interpretation corpus through 2–3 progressive AHJ pilots. The insurer underwriting in 2027 is the inflection signal — the bet has a clean go/no-go.
2. Avoid the generative-BIM concept-tool category as a standalone bet. Snaptrude and peers are valuable but bounded. The simple-vs-complex split makes them acquired-base, $200M–$700M, with Autodesk as the natural buyer (Spacemaker→Autodesk precedent). Capital deployed here is pricing in the bear, not the bull. Better risk-adjusted return: deploy into Archetype 1 or Archetype 3 instead.
3. Watch Speckle's commercial trajectory in 2026–2027 as the gating signal for Archetype 3. If Speckle is acquired by Autodesk or stalls in the commercial tier, the post-Speckle per-project commercial-first lane opens and is investable in late 2026 / early 2027. If Speckle locks in independence at $1–2B, the lane is closed and Archetype 3 becomes a worse bet.
4. If you are Autodesk: ship the credibly open project-graph layer voluntarily, at margin cost. The counter-positioning trap (chapter 03; chapter 13's loud claim 3) tightens with every dollar of AI investment that lands on top of seat licensing. The cheapest defense is the one Autodesk has not yet taken — concede the API-perimeter battle and re-anchor on the accountability-stamping layer that Priya's analysis says is structurally untouched at H2. Per-project pricing on the project-graph layer; per-seat pricing reserved for documentation and stamping. The pricing-model fracture is real but smaller than the cannibalization that the seat-only path delivers over 5 years.
5. If you are Procore: launch a consumption-tier mid-market offering by 2027 even at partial cannibalization. The DecagonCo-AEC archetype eats the mid-market wedge cleanly otherwise. The cannibalization to upmarket is bounded because consumption-priced mid-market customers are not the upmarket-volume customers; the lost revenue is from a customer base Procore is not currently capturing. Defending the upmarket pricing model by leaving mid-market unserved is the worst available defense.

The framework's closing prediction: AEC software in 2029 looks materially different from 2026 not because Revit was displaced, but because (a) per-outcome and per-project pricing has eaten the typology edges and the mid-market wedge, (b) the API perimeter has been ceded to an open or commercial-first project-graph layer, and (c) the accountability-stamping layer has emerged as the structurally durable piece of the value chain — and the entity that owns it (Autodesk if it pivots; an insurer-underwritten Evaluator-Power player if it does not) holds the most defensible position in the industry. The seat-licensed dollar-weighted middle is not displaced; it is squeezed harder than its pricing model can absorb.