heyarnoux.

For CMOs

The Marketing Function Is Becoming an Engineering Function

Uber embedded AI engineers inside every business function and marketing QA went from two weeks to under an hour. Here is the case for running your marketing org the same way, and what changes for you personally.

Uber's CTO published a number every CMO should sit with for a minute. After the company embedded AI engineers directly inside its business teams, marketing QA that took two weeks now takes under an hour.

That result did not come from buying a tool. It came from an org design decision. Praveen Neppalli Naga, Uber's CTO, described the mechanism. Uber picked around 30 of its most AI-proficient engineers and paired each one with a domain expert inside a business function. Finance, legal, marketing, support, HR. They call them agentic pods, and they ran 16 of them across 16 business functions in two months.

I have spent the last years building exactly this inside marketing orgs, more than 20 times, from seed-stage to a €2.5B enterprise. So this piece makes the argument directly. Marketing should be run like an engineering function, with engineers embedded in the build. And it walks through what that means for how you, the CMO, run your week.

What Uber actually did

Each pod gets two weeks. The engineer starts by shadowing the domain expert, watching the real workflow step by step before building anything. Then they prioritize the automation opportunities by scale and impact, build agents against real systems with the person who does the job sitting next to them, validate with others doing the same work, and ship on day ten.

01PairOne engineer, one domain expert, one pod
02UnderstandShadow the real workflow for two days before building
03IdentifyRank the opportunities by scale, repetition, impact
04BuildAgents and workflows against real systems, days four and five
05ValidateTest with others who do the same job, measure the impact
06ShipDay ten. In production, in the hands of the team.
Uber's agentic pod cycle. Two weeks from pairing to production.

The backdrop makes the pods possible. By Uber's own count, 99% of its engineers use AI tools, more than 70% of pull requests are attributed to local or cloud agents, and engineers have built over 2,500 agent skills covering the development lifecycle. The pods take that maturity and walk it into the business.

The results are the part worth reading twice. Capital allocation analysis dropped from 15 hours to 30 minutes. Financial pacing reports that took two days now take ten. A library of 9,000 manually built support workflows was replaced with self-service automation. And the marketing one, QA on campaigns going from two weeks to under an hour.

2wks → 1hrmarketing QA
after one pod
70%+of pull requests
attributed to agents
16 podsacross 16 functions
in two months
Uber's published numbers, per its CTO
The engineers moved to the work. The domain experts stayed in the loop. Production in ten days.

Read that structure again and you will recognize it. One operator who owns the outcome. One person who knows the systems. One engineer who builds. It is the same three-seat unit I map in the marketing engineering org chart, discovered independently by a company with 30,000 employees and a share price.

The hubShared brain + standards. Thin, purely enabling. Every function pulls from it.
Marketing
OperatorOwns the problem + the outcome
Ops leadOwns systems, data, definitions
GTM engineerBuilds. The hard seat to fill.
Sales
OperatorOwns the problem + the outcome
Ops leadOwns systems, data, definitions
GTM engineerBuilds. The hard seat to fill.
Customer success
OperatorOwns the problem + the outcome
Ops leadOwns systems, data, definitions
GTM engineerBuilds. The hard seat to fill.
The three-seat unit repeats per function. The hub stays thin.

Why marketing is the function this hits first

AI collapsed the cost of building software. What it left scarce is distribution, getting in front of the right people and closing them. That moved the moat onto the marketing floor, and the market has already priced it. Postings for the GTM engineer grew 205% between 2024 and 2025 per Bloomberry's analysis of 1,000+ listings, with a $127,500 median and Vercel and OpenAI paying around $250,000 for the title.

Marketing is also unusually exposed to exactly what agents do well. The function runs on repeatable text-and-data work. Enrichment, scoring, QA, versioning, reporting, list building, personalization. The research says the build is moving in-house, with GTM engineering becoming the technical execution layer while traditional ops scopes contract. RevOps keeps process and governance. The engineer builds the new engines on top.

So the honest framing for a CMO is capacity planning. Every quarter without build capacity inside your org, your workflows stay manual while a competitor's marketing QA drops to an hour. Uber just published what the gap looks like.

Proof from a marketing org

If Uber feels far from your world, take a pure marketing example. Backbase, a €2.5B fintech, rebuilt its go-to-market this way and its CMO Tim Rutten has been open about the results. Double the pipeline on roughly 25% less budget. A 35-person marketing team against a book north of $350M in ARR. An in-house signal engine scanning 3,000 target banks for about $500 a week, replacing a six-figure tool.

Notice what both stories share. Neither bought its way there with software. Both changed who sits inside the marketing org and how the work ships.

What this looks like running

Talking about engines is cheap, so here is one running. I recorded a walkthrough of a working GTM OS, the shared brain under a marketing org run this way. Signals, workflows, agents and reporting in one system a three-seat unit operates.

What a working GTM OS looks like, recorded from a live build

Prefer to click through it yourself? The same GTM OS is live as an interactive demo.

Walk through the GTM OS →

What changes for you

Running marketing like an engineering function is mostly a change in what the CMO personally rewards, funds and reviews. Six concrete moves.

Budget
Fund build capacity before tools. Backbase doubled pipeline on ~25% less budget by building in-house. Every six-figure platform on your stack is now a question. Could one engineer and $500 a week do this better?
Hiring
Your next hire is a GTM engineer. Install the three-seat unit per function. Operator, ops lead, engineer. The engineer seat is scarce and priced accordingly, so decide early whether you hire, borrow or grow one.
Cadence
Two-week ships, working demos over decks. Uber's pods go from pairing to production in ten days. If your team's unit of progress is still the campaign or the quarter, that is the first thing to change.
Data
Own the semantic layer. The written dictionary of what your business means by MQL, account, customer. Without it the smartest agent gets your world confidently wrong. On my biggest rebuild the first nine months went into cleaning the CRM.
Governance
Keep one human line. Someone sets the parameters and approves anything that matters before it goes out. Uber paired every engineer with a domain expert for the same reason. The accountability stays with people.
Your week
Review running systems, then their numbers. The engineering-org CMO opens the dashboard the agents act on, watches an engine demo and asks what shipped. The slide deck comes last, if at all.
The CMO's side of the deal

How to start without a reorg

You do not need a transformation program or a steering committee. You need one pod. Uber proved the pattern at 16 functions in two months. You need it once.

01Pick the worst workflowOne painful, repeatable process with a number attached
02Run one podOne engineer + one operator, shadow first, ship in two weeks
03Publish the numberBefore and after, internally. Let the org chase the evidence.
One pod is the whole pilot

Then the flywheel is sociology. When one team's two-week task takes an hour, every other team asks for its pod. That is how Uber scaled it and how the lighthouse rollout works in every org I have built this in.

The uncomfortable part

The hard question this raises for a CMO is about identity. The job description quietly rewrote itself. You are now the owner of a production system that generates demand, with brand and creative as inputs to it. The CMOs who thrive in that world read a pacing dashboard as fluently as a brand deck, sit engineers at the table when the plan is made and treat "we shipped" as the weekly heartbeat.

The role is bigger, honestly. An engineering-run marketing function compounds. Every workflow you automate stays automated, every definition you clean stays clean, every engine keeps running while your team thinks. Channel-team marketing starts every quarter from zero. That compounding is what Uber bought with 30 engineers and what Backbase bought with one org chart.

Building got cheap. Distribution is the moat. The companies treating marketing as the engineering function that owns it will set the pace, and the published numbers say they already do.

Sources

  • Uber's agentic pods, 99% AI tool adoption, 70%+ of pull requests from agents, 2,500 agent skills · Praveen Neppalli Naga (CTO, Uber)
  • Pod structure, 16 pods in two months, marketing QA two weeks to under an hour, capital allocation 15 hours to 30 minutes · The State of AI
  • GTM engineer job growth (205%) and $127,500 median, from an analysis of 1,000+ listings · Bloomberry
  • GTM engineer salaries at Vercel and OpenAI · eMarketer
  • The build moving in-house, GTM engineering as the execution layer · MarTech
  • RevOps versus GTM engineering · Factors.ai
  • Backbase results (double pipeline on ~25% less budget, the in-house signal engine) · FullFunnel case study
  • Video walkthrough of a working GTM OS · David Arnoux on YouTube

Build yours

Want a pod inside your marketing org?

I install the three-seat unit and the shared brain, one department at a time, and hand it over working. Walk through a live GTM OS, or book a 30 minute fit call.