Market memo · 2026
The Transformation Stack
How we read the market, where we play, and why Layer 4 is unclaimed.
TL;DR
A new market is forming around the AI-driven transformation of traditional businesses. It is being built by four kinds of buyers — capital pools, AI-native holdcos, AI-lab joint ventures, and corporate transformation teams — and is being served by a fragmented stack of vendors that mostly attack the easy layers.
The hardest and most valuable layer in the stack — turning observed real-world work into structured specifications that an AI engineer can build against — is essentially unclaimed. That layer is where Aperture plays.
What is happening
Three things are converging right now that have not converged before, and the convergence is what creates the opportunity.
The first is on the capital side. Top-tier VC firms have begun deploying capital into a strategy that looks more like private equity than venture. They are funding AI-native operators that acquire traditional services businesses — call centers, accounting firms, property managers, legal services — and transform them with proprietary AI tooling. General Catalyst alone has dedicated $1.5B to this thesis and co-created at least ten such platforms. Thrive has launched Thrive Holdings, a $1B+ permanent vehicle with the same mandate.
The second is on the AI lab side. In May 2026, Anthropic announced a $1.5B joint venture with Blackstone, Hellman & Friedman, and Goldman Sachs to deploy Claude into mid-market PE portfolio companies. Hours earlier, OpenAI announced DeployCo, a $10B vehicle with TPG, Brookfield, Bain, and Advent. Both are explicit forward-deployed-engineer plays. Both target the ~2,000 portfolio companies in their PE backers' books as the initial deployment surface.
The third is the technical unlock. Frontier models can now read 20-year-old ABAP and figure out what business problem it was written to solve. They can map fields across heterogeneous schemas. They can infer business logic from observed system behavior when documentation is missing. The category of work that has resisted automation for thirty years — the judgment-heavy, code-comprehension-heavy work that consultants do at $300/hour — is suddenly tractable.
Read these three together: large pools of capital, a delivery model that bypasses traditional consultancies, and an AI capability that can actually do the work. The most important date on the calendar is October 2027, when SAP ends standard support for ECC. Roughly 10,000 ECC environments globally will be forced to migrate. That deadline is the forcing function.
The five-layer stack
The single most useful mental model for placing every player in this market is to think of AI-driven enterprise transformation as a stack of five layers.
Layer 1 — Capital & Operator Vehicle. Who funds the transformation and who owns the resulting business. Traditional PE, AI-native operators, AI-lab JVs, and in-house transformation budgets at Fortune 500 corporates. This is the buyer for everything above them in the stack.
Layer 2 — Strategic Discovery. The pre-deal and post-deal-close work that decides what to buy and what to transform. Today this is dominated by McKinsey, Bain, BCG, expert networks, and a handful of AI diligence tools. It is largely document-based.
Layer 3 — System Migration. The technical conversion of legacy systems to modern ones. ERP transformation, code modernization. Heavily technical, code-and-data-centric, freshly unlocked by frontier-model code comprehension.
Layer 4 — Workflow Discovery & Specification. The translation of how work actually happens in a real company into structured artifacts that an engineer can act on. This means observing humans across screens and systems, segmenting the chaos into discrete workflows, abstracting them into typed specifications, scoring them for automation potential, and emitting machine-readable specs that downstream engineers can build agents from. Today this layer is partially served by process mining and task mining — but none of those tools emit specifications. They emit dashboards. The handoff to the engineer is still manual, slow, and lossy.
Layer 5 — Build & Run. The actual implementation of agents, automations, and AI-driven workflows inside the customer's systems, plus the ongoing operations layer once they're live. Forward-deployed engineers at frontier labs and the new AI engineering practices at Big 4 firms sit here.
The pattern, once you see it: most well-funded companies in the market are clustered in Layers 1, 3, and 5. Layer 2 is dominated by incumbents who are slow to adopt AI. Layer 4 is barely populated.
Buyer archetypes
Layer 1 is not a single buyer. It is four different kinds of buyer with four different procurement motions. Each motion has a different cap-table, a different deployment surface, and a different expectation about what a vendor delivers.
Internal agent engineering teams. A new function emerging at every Fortune 1000 — internal teams chartered to deploy AI agents inside the company's own workflows. Job titles vary (AI automation engineer, agent PM, applied AI engineer) but the mandate is consistent: figure out which workflows agents can take on, wire the integrations, operate the output. They're under headcount + budget pressure to ship fast and are emerging as a named buyer with explicit tooling budgets. A particularly acute version of this archetype is the AI-native roll-up operator — companies founded in the last 2–4 years to acquire traditional services businesses and apply the same agent-engineering playbook portfolio-wide. The internal-agent-engineering function is the broader market; roll-ups are one concentrated deployment surface.
AI-lab joint ventures. The most concentrated buyer in the world right now. Each JV represents $1.5B–$10B of capital and access to thousands of portfolio companies as captive deployment surface. Both major JVs have explicitly committed to a partner ecosystem rather than building everything internally.
Internal corporate transformation teams. The largest aggregate footprint of in-flight transformation work, with procurement governed by formal vendor review and internal stakeholder alignment. Every Fortune 1000 has an internal team trying to do AI transformation on its own ERP, CRM, and operational stack.
Traditional SIs pivoting to FDE. Accenture, EY, KPMG, Deloitte, IBM, plus regional equivalents. Spinning up forward-deployed engineering practices as a defensive measure. These practices look for tooling that lets them deliver FDE-style outcomes within the economics of a traditional consulting engagement.
Each archetype is buying transformation outcomes; what differs is the deliverable shape, the procurement timeline, and who inside the organization signs the work.
Where competition sits
Place every competitor on the stack and the gap becomes obvious.
Layer 2 (diligence): AI diligence tools, expert networks, legal-tech AI. All document-centric. None observe live work or produce workflow specs.
Layer 3 (migration): ECC→S/4HANA specialists, broader pre/post-migration platforms, assurance and governance layers, the SAP migration practices of every Big 4 firm.
Layer 4 (where Aperture plays): event-log process mining, desktop task mining, tool-by-tool semantic search. None of these produce FDE-ready specifications. None integrate captured behavior into eval sets for downstream agent building. None have a workflow taxonomy that compresses across customers.
Layer 5 (build & run): AI-lab JVs, frontier-lab forward-deployed engineers, the AI engineering practices being stood up at every Big 4 firm.
The gap that matters is Layer 4. Every other layer has multiple well-funded incumbents. Layer 4 has older-generation tools that produce dashboards, not specifications, and that were architected for an era when AI couldn't read code. There is no startup that has built a Layer 4 product purpose-designed to feed Layer 5 FDEs.
The reason this gap exists is structural: process mining grew up serving Six Sigma transformation programs in the 2010s, where the deliverable was a PowerPoint to the COO. Today's deliverable is a working agent. The category needs to be rebuilt for the new buyer.
Closing
The market we are building into is being created in real time by capital and AI labs that are rewriting how enterprise IT spend gets allocated. Most of the visible activity is at Layer 1, Layer 3, and Layer 5. The middle of the stack — Layer 4, the place where observed work becomes machine-buildable specification — is unclaimed. That gap exists because the tools that grew up there were built for a different buyer and a different deliverable.
If this maps to a problem you're working on, talk to us.