← narwal.one/Second Brain
SecondBrain
Ask the Brain
Index/Conceptupdated Tue Oct 20 2026 08:00:00 GMT+0800 (Philippine Standard Time)

Agentic Pods

uberagentic-transformationforward-deployedinternally-deployedworkflow-automationpodsbusiness-functions

Agentic Pods

Uber's July 2026 operational recipe for extending agentic AI beyond engineering into every business function — 2-week pods pairing one of the company's most-AI-proficient engineers with one domain expert from a target function. Announced by Uber CTO Praveen Napali on X (reported by Nathaniel Whittemore in How to Help People Thrive with AI (AI Daily Brief)).

The vault's cleanest working example of "internally deployed vibe coders" — Whittemore's 2026 prediction for how forward-deployed engineering practice moves inside an enterprise.

The recipe

  • ~30 handpicked engineers, each "with deep knowledge of Uber's systems" + AI-proficient
  • Each paired with a domain expert from a business function (finance, legal, operations, marketing, customer support, HR, procurement)
  • 2-week timebox per pod

The 10-day cadence:

Day(s) Activity
1–2 Shadow the expert. Observe every step; document workflows; ask questions; build intuition.
3 Prioritise opportunities by scale × repetition × business impact × data availability.
4–5 Build a working agent alongside the person doing the job.
6–9 Validate with several others performing the same work. Does it generalise? Does it actually make their job better?
10 Ship.

Napali's headline productivity results (16 pods in 2 months)

Workflow Before After
Capital allocation across 150 cities 15 hours 30 minutes
Financial pacing reports 2 days 10 minutes
Marketing web quality assurance 2 weeks 50 minutes
Support workflow creation ~9,000 manual workflows Self-service automation

The engineering-side context Napali provides at the top of the thread: "Today, 99% of our engineers use AI tools. More than 70% of pull requests are attributed to local or cloud agents, and our engineers have built 2,500-plus agent skills across the software development life cycle." → Uber built the engineering-side capability first; pods generalise that capability across the enterprise.

The load-bearing quote

Napali's own summary of what surprised the Uber team the most:

"It was how quickly engineers embedded in unfamiliar domains uncovered opportunities that had been hiding in plain sight. The biggest wins rarely come from automating one task. They come from rethinking an entire workflow. Once you redesign the workflow around AI, you often eliminate handoffs, remove unnecessary approvals, replace legacy tooling, reduce vendor spend, and dramatically accelerate decision-making. The workflow becomes the unit of automation, not the individual task. The most impactful agent skills cut across teams, orgs, functions, tools, and systems."

And on the discovery mechanism:

"The best AI opportunities are rarely visible from the outside. You discover them by sitting next to the people doing the work, understanding every friction point, and building with them, not for them."

Why this is the load-bearing case study

The vault has been carrying multiple altitude-versions of the tasks → responsibilities shift:

  • Individual (Ryberg)
  • Team (Brovich hyperconvergence pods)
  • Org (Zehren Exponential IT)
  • Firm (Alströmer YC AI-native services)

Napali's pods deliver a cross-functional workflow altitude — the workflow itself is the unit of automation, not the individual task inside it. This is the operational move Brovich pointed toward with hyperconvergence, but Uber has now shown what the day-by-day cadence looks like inside a large enterprise.

Uber is also now "forming a dedicated team to scale this further" — pods graduate from a one-off transformation programme into a permanent enterprise function (adjacent to IT Ops Model B — embedded pods — becoming institutional infrastructure).

Whittemore's second-order framing

The 2-week results are impressive but not the payoff. The real payoff comes over months as the business people themselves — newly influenced by agentic work with an embedded engineer — start to think differently about what they should be doing at all:

"While the engineers might help the financial pacing reports move from 2 days to 10 minutes, it is in many cases going to be the business folks, newly influenced by these agent techniques and maybe even building and working with some agents themselves, who figure out the best way to spend the other 1 day, 23 hours, and 50 minutes. And in many, if not most cases, that won't be doing more of the same work. It will likely be doing new work, orthogonal work, work that was always dreamed but never possible before."

This is the enterprise-scale version of Whittemore's central "use AI for things you can't do, not just things you can already do" insight — pods produce productivity gains, but the reinvestment of those gains is what transforms the business.

Where this connects to the vault

  • Tasks to Responsibilities Shift — the cross-functional workflow altitude; a new row in the four-altitude ladder (individual / team / org / firm) that also includes the workflow-as-unit-of-automation.
  • IT Ops Models (A B C) — pods are the Brovich Model B (embedded pods) made operational; Uber's "dedicated team to scale this further" is Model B + Model C (pods on top of platform).
  • Software Factory — a pod is a factory-assembly unit that produces agent skills for a specific domain workflow. Uber's 2,500+ agent skills are the factory output; the pods are the assembly floor.
  • Skill Engineering — pods are a skill-engineering-as-discipline operational cadence: the 2-week rhythm is the shortest working cycle from "shadow the expert" to "ship a skill."
  • Tacit Knowledge — pods are Route-4 (social/embodied capture) — the engineer sits next to the domain expert and encodes what the expert knows-but-can-tell-you-only-if-you-ask. Complements Route-3 (video / keystroke) capture.
  • Company Brain + What an Enterprise Context Layer Is (Prukalpa) — the 2,500+ agent skills accumulating across 16 functions in 2 months is the substrate-and-skills layer of an enterprise context layer being built from below.
  • FOBO (Fear of Becoming Obsolete) — pods are a working answer to the "someone using AI will take your job" framing: the domain expert is co-building the agent that will run their workflow, which is the opposite of being replaced by it. Rebrand-able for MIH audiences as the "be the person the pod is built with, not around" frame.
  • Verification Tax — Uber's 70%-agent-attributed PRs + 2,500+ skills prove the tax is real and operable at scale; Napali doesn't dig into their verification mechanism but the fact that pods can validate + ship in a 10-day cadence is scale evidence the tax can be paid on rhythm.
  • Reverse Information Paradox — pods keep the "opportunities hiding in plain sight" + workflow redesigns inside Uber; the intelligence-exhaust from the pods programme accrues to Uber's operational capability, not the vendor.

Connects to your work

For P&G Manila IT (MIH) and the Manila IT podcast series:

  • The pods pattern is directly imitable at MIH scale. 30 AI-proficient MIH engineers × 30 domain experts (across finance / R&D / operations / marketing / customer / HR / procurement / product / plant / supply chain) × 2-week timebox × 16-functions-in-2-months would be a 90-day transformation programme that generates a MIH-specific skill catalogue.
  • The "sit next to the person doing the work" discovery mechanism is the exact opposite of process-diagram-driven transformation and the exact match for Jejurikar's "culture as the gating factor" — pods are a cultural mechanism first, a technology mechanism second.
  • "The workflow becomes the unit of automation, not the individual task" is the vault's cleanest one-line summary of what agentic transformation means at enterprise scale — usable as a podcast title / opening quote.
  • The two-week cadence matches MIH's existing biweekly sprint structure; the constraint is not process, it's identifying the right ~30 AI-proficient engineers to seed the programme.

Autonomous-mode note

Single-source (Napali on X + Whittemore's reading). First substantive vault touch of Uber-as-transformation-case-study; second substantive touch of Uber overall (after the Token Scarcity Uber-$1,500/mo-cap datapoint). If a customer case study or Latent Space long-form deep-dive on the Uber pods programme lands, the confidence rises considerably. Until then, this concept page is reporting what Napali claims, cross-linked to the vault's existing tasks-to-responsibilities architecture. Kept the framing careful: "Napali's numbers" not "the numbers", no independent verification yet.

2026-10-20 — Where pods sit on the champions spectrum (per How to Go from AI Power User to AI Leader (AI Daily Brief))

Nufar Gaspar's ambassador → ambassador-builder → internal FDE spectrum gives the pods pattern a taxonomy: Uber's engineer × domain-expert pairs are the internal-FDE end, run as a programme rather than as individual champions. Her nomination signals (manager gives them the time; peers already come to them; business knowledge, not new joiners) and the "day after" sponsor model are the selection and governance layer the Napali thread doesn't describe — see AI Champions Network. Her "simplest thing that solves it" ladder (chat setup → skill → automation → agent) is also a useful brake on pods defaulting to full agents.

Sources