There is a day, in every cycle of enterprise software, when the layer where value accumulates moves. Today was one of those days.

a16z just published From System of Record to System of Intelligence. The byline is Gio Ahern, Steph Zhang, and Alex Immerman — all growth partners at a16z. The short version: Salesforce and HubSpot aren’t going anywhere, but they become what Facebook’s friend graph became after the news feed shipped. A data layer, not the thing the user opens every morning. Your CRM is infrastructure now. The value moved up to the reasoning layer above it.

That’s the easy half of the shift. The uncomfortable half is what happens if you run technology for a company in the 80-to-2,000-employee range and you’re not sitting in Silicon Valley. Anthropic is not sending you a Forward Deployed Engineer. Neither is Google. The reasoning layer either gets built by you, or it gets built — two years later, at eight figures — by a Big 4 firm invoicing you to discover what your team already knew.

Orchestration work is the product now

That’s the function IQ Source occupies, and the reason the service is framed exactly this way.

We don’t sell an agent. We don’t sell a clever prompt, and we don’t sell a one-off integration. What we sell is the Forward Deployed Engineer function, packaged for mid-market companies that aren’t on Salesforce and that don’t have the budget — or the need — to hire McKinsey. Two chained phases:

AI Agent Maestro is the discovery half. Two months. The same work an Anthropic FDE would do in their first eight weeks inside a Fortune 500 customer: a process map grounded in actual operating reality (not the SOP), an AI Opportunity Score across processes, identification of the workflows where the reasoning layer delivers measurable economic value, and a Go/No-Go gate at the end. The service page lays out the exact deliverables.

Technology Partner is the build half. Here is where the work Aakash Gupta describes as “the other 80%” actually happens: integrations into your HubSpot, your legacy Odoo or SAP, the two spreadsheets nobody wants to migrate; eval design; hallucination guardrails; coordination with security and legal; and the change management discipline Allie Miller correctly names as the part everyone underestimates. The Technology Partner page has the scope.

That’s the thesis. Everything else in this piece is why, in May 2026, the evidence stacked up enough to defend it publicly.

What a16z actually argued (and why it’s confirmation, not novelty)

For twenty years, the winner in enterprise GTM software was whoever owned the database. Salesforce and HubSpot accumulated two decades of operating context — every call note, every price precedent, every contact, every observation about why a deal stalled. That accumulated context built the switching cost. As Alex Rampell of a16z put it a few years ago, users became “hostages, not customers.”

AI agents change the gravity. An agent doesn’t need a drag-and-drop pipeline view. It needs structured data it can read and write with low friction. The CRM, from an agent’s perspective, is a trusted, well-integrated database with a decade of curated context — but a database nonetheless. The opinionated workflow, the daily prioritization decision, the question of which account to call this morning — that has moved up.

The line from the essay worth keeping verbatim: “The reasoning layer that sits above the database, and that increasingly treats the database as infrastructure, is where a new generation of companies is being built, and it’s where the majority of the next decade’s enterprise value of GTM software will end up.”

Two findings worth holding onto. First, CRM usage went up after AI tools were adopted, not down: agents that listen to calls and write structured notes back into the system have made the data richer than it ever was. The database doesn’t die; it stops being where the user looks first. Second, software has historically been 5–10% of total GTM spend; the rest is payroll. The reasoning layer, for the first time, opens a path for software to grow ROI without cannibalizing headcount.

This isn’t novel. I made the same argument three weeks ago from a different angle: if the model commoditizes and context gets pulled inside the tools, the moat lives in the workflow. The workflow is what a16z is now calling “the system of intelligence.”

The Forward Deployed Engineer explosion is the proof

If this were only a VC argument, it would be easy to dismiss. It isn’t. The labor market of the last two weeks is the evidence.

Aaron Levie, CEO of Box, argued yesterday that university career services should be teaching students what a Forward Deployed Engineer is and how to land one of those jobs. His thesis: hundreds — maybe thousands — of tech companies are hiring for this role, and supply is well below demand. Nader Dabit, from Cognition, put it more plainly in a separate thread: OpenAI, Anthropic, Google, and Cognition are all hiring aggressively.

Aakash Gupta wrote the most useful structural read. His piece yesterday documents that Google is hiring hundreds of FDEs for Gemini, copying — verbatim — the model that took Palantir from $16B in 2020 to $325B today. Aakash’s conclusion, in its most direct form:

The model itself is roughly 20% of an enterprise AI deployment. Integration, eval design, hallucination guardrails, security review, change management, and getting employees to actually use the thing is 80%. Selling the model alone is selling a fifth of the problem.

Two economic consequences fall out of this. First, pure software runs at 75–85% gross margin; professional services runs at 30–40%. When Google announces it will hire its own FDEs, it’s telling the market that enterprise AI revenue arrives with a services tail attached, and that tail compresses the multiple every analyst was modeling. Second, Accenture, Deloitte, and McKinsey built a combined $90B business by being the deployment layer on top of cloud and SaaS. Thomas Kurian — Google Cloud’s CEO, who came from Oracle, where consulting was always the second engine attached to every license — is internalizing that layer.

For you, running technology at a mid-market company, this isn’t Silicon Valley news. It’s the signal that the layer where value accumulates moved, and the companies defending it are the ones that can hire 200 internal FDEs. You can’t. You need a different route.

Allie K. Miller’s caveat is the one you can’t skip

Allie K. Miller — ex-Amazon, ex-IBM, one of the most-read voices in enterprise AI — adds the necessary correction.

Her piece this week names the expensive mistake executives infatuated with hiring three FDEs and declaring transformation are making: “You can have the sleekest multi-agent orchestrations and still have the majority of your employee base hating AI, avoiding AI, and distrusting leadership decisions on AI.”

Transformation compounds from tech + people enablement + process reinvention. Prioritize only the first and you stall.

That is exactly the reason the AI Agent Maestro phase exists before anything gets built. Before a single integration is moved, the actual process gets mapped (not the documented one), and decisions are made on which tasks to automate, which ones to redesign, and which ones to kill outright, with the reason for each. Skip this phase and what you build works technically and fails operationally — the ERP playbook of two decades ago, replaying. The technology isn’t the problem. Getting people to actually use it is.

What to do this Friday if you run technology at a mid-market company

Three concrete moves, in increasing order of commitment.

One. Look at your stack this afternoon. List the systems of record your operation actually uses every day: CRM, ERP, support tool, accounting system, the two spreadsheets nobody wants to migrate. Mark which ones have useful APIs and which don’t. That list is your system-of-record inventory. The reasoning layer you’ll build sits on top of it. Without the list, any conversation about agents is vapor.

Two. Identify a single process where the cost of “the other 80%” — integration, evals, change management — is affordable and the return is defensible inside three months. One process. Not ten. In B2B mid-market, the pattern with the strongest ROI is usually lead triage + account research + follow-up drafting, because it touches three systems (CRM, inbox, calendar) and the return shows up as closed quota, not as time saved.

Three. Decide who is going to do the FDE work for you. There are three honest options: hire internally (expensive, slow, almost impossible to close in LatAm with the right profile), engage a Big 4 (eighteen months, eight figures, the discovery deliverable lands after the build window has already shifted), or work with a regional partner who already lives in this stack, already has the discovery and build discipline, and sizes the engagement to your actual scale. That third lane is the one IQ Source occupies.

If you want to see what discovery looks like, the AI Agent Maestro page has the deliverables, the two-month timeline, and the Go/No-Go gate. If discovery is already done and you need the build phase, the Technology Partner page has the scope.

Your CRM isn’t going anywhere. Nobody is taking it from you. But the question your next board meeting will surface is whether you own the reasoning layer above it — or whether the opportunity you could have captured this quarter ends up on a Big 4 invoice two years from now.