AI & Social Acceleration: Why Are We Faster Yet Falling Behind? – Cat Swetel (Talk Outline)

Cat Swetel (GM of Nubank’s foundation/infrastructure unit, with a background in science & technology policy) uses sociologist Hartmut Rosa’s acceleration theory and Little’s Law to explain the paradox: if AI makes us more productive, why is everyone exhausted? Her answer — pace mismatch, premature materialization, rising arrival rates with flat departure rates, a shrinking present, and “frenetic standstill” — leads to a very human prescription: intuition, embodiment, and questioning perpetual growth.

1. Framing & Speaker

• Cat Swetel, GM of the foundation business unit at Nubank — a Latin-American bank with ~130 million customers — running infrastructure and platform tooling, and stewarding the Datomic (append-only) database (colleague Jen can go deeper; Cat won’t).
• Lens for this talk: her academic background — a master’s in science & technology policy, studying the impact of technological innovation on society. (Playful aside: “Leo rising, Taurus sun and moon.“)

2. “How Are You Feeling?” — The New Normal

• Audience is tired. Everyone has a CEO “back from Jeff Bezos camp” mandating AI everywhere, even a token-burn quota — bizarre given “tighten the purse strings” a year ago.
• Hard truth: the exhaustion is not temporary. People say “we’ll just push through AI adoption, then things go back to normal” — but this is the new normal; there is no going back (like the internet making work follow you home).
• Takeaway #1 (even if you leave now): it’s not temporary and there’s no “pushing through” — you must find a way to cope.
• Temporary morale boosters (hackathons, cute t-shirts) won’t power you through an entire era of computing — you need sustainable coping.

3. The Central Question

• If AI makes us more productive, why are we all so tired? Shouldn’t we have more time to rest / do energizing things? Nobody in the room is “chilling out.” Counterintuitive: we offloaded tasks to AI, yet we’re exhausted.

4. Hartmut Rosa’s Acceleration Model

• Sociologist Hartmut Rosa: modernity is accelerating, and three forms reinforce each other:
  • Technological acceleration →
  • Acceleration of social change →
  • Acceleration of the pace of life → (loops back).
• Example: connectivity doesn’t make life less busy — you get paged on your phone (Cat carries a large phone as an executive escalation point for critical incidents; “always on”).
• Social norms change with acceleration. Fast food wasn’t invented so families could enjoy dinner together — it lets you grab food on the way to the next activity, making life denser (multitasking even while eating) — “weird and a little sad.”

4.1 Not everything accelerates — pace mismatch

• The feeling that “everything is getting faster” isn’t fully true. With AI:
  • Making a baby still takes 9 months (no 10×).
  • Recovering from a cold/flu still takes the same time.
• Some paces are physically in us and don’t change — and it gets interesting when they collide with social change:
  • Job tenure used to span a career (his dad retired from his first job); now big-tech average tenure is ~14–18 months — so companies offer perks like “we’ll pay you to freeze your eggs” because biology (9 months) doesn’t match tenure.
  • Traffic: everyone getting a car is a huge individual convenience, but on average everyone arrives slower (congestion). Feels faster, but some things stay the same and some get slower — a pace mismatch.

5. Humans Have Rhythm; AI Does Not

• Humans have a circadian rhythm — energy peaks and troughs, sleep. AI is just continuous.
• Cross-team coordination is hard because each team has its own rhythm (stand-up this day, planning that time); combining two rhythms gets “janky.” AI has no rhythm at all — so integrating across it is hard.

5.1 Agentic development and premature materialization

• Agentic dev materializes the prompt into an actual thing quickly. Before, an idea went through user stories, sketching, architecture, design — many points to diverge and converge / iterate.
• Now there’s strong pressure toward convergence with no respect for our rhythm.
• A materialized thing is harder to argue with — “it already exists, I saw the page, I interacted with it.” Your brain stops generating its own design and instead compares to the existing one.
• Materialized designs easily become a commitment — we have a strong aversion to turning things off (it feels like a loss), while our ability to turn things on has increased. Agents rarely say “we’re done, let’s turn it off,” so we accumulate things to care for.
• Lesson: our work with AI needs to (re)learn what humans learned — take much smaller steps. Cheap-to-make doesn’t mean it’s best to jump to a highly materialized design; disparate people materializing separately are hard to merge.

6. Little’s Law — Arrival vs. Departure Rate

• Little’s Law (known intuitively): work arrives at a rate and is delivered at a rate. If arrival ≈ departure, work-in-process is steady and lead time is steady.
• If arrival rate shoots up while departure rate stays flat, WIP grows and each item takes longer to actually complete (deliver to a customer — not chuck to another team, which is just someone else’s arrival rate).
• Today there’s huge emphasis on arrival rate (agentic prototypes, “we built it so fast!”), but research shows departure rates are roughly the same — lots of stuff made, little extra value shipped. We accumulate things to care for without delivering more value — that’s when we feel the acceleration / overwhelm.

6.1 Departure rates can even get worse

• Recent GitHub incidents, AWS region degradations — “things seem to be getting shittier,” possible quality issues making shipping harder.
• Regulators watch the news; in regulated industries (like banking) there are new requirements for handling outages of critical infrastructure — further inhibiting departure rate.
• So technical acceleration leads to an acceleration in the experience of work, not in value delivered.

7. Frenetic Standstill & the Shrinking Present

• We accumulate technical debt faster than value — we don’t see the usual value indicators of a big innovation wave. Hartmut Rosa’s term: “frenetic standstill” — doing a lot without traction. Not “AI sucks,” just that we’re early and not yet capturing the benefits.

7.1 The concept of “the present” contracts

• Ask “what are you working on right now?” — the time horizon of the answer varies: days, quarters, or quarters-to-years. We define “now” by how far our expectations for our actions extend — the past = “I can’t do anything about it,” the future = “I don’t know what will happen.”
• Under AI, new things launch constantly → industry unpredictability → the present condenses to weeks/months even at CEO/CTO level (AI war rooms meeting daily for two weeks).
• Cascade: if C-level’s present is weeks, the next level down operates in weeks, then days, then hours — collapsing the normal nested strategy horizons. With no expectations extending into the future, the only “safe” action is very proximate and probably not impactful → burning tokens, accumulating low-impact things → frenetic standstill.

8. Now What? — The Human Value

• “16 more minutes of depression” — then the pivot: what value do humans bring?

8.1 Intuition

• Machines don’t have intuition. Technical people are trained on ”the truth is in production, show me the data, no handshake deals.” Cat went the opposite direction (“turned into a crystal girly” — essential oils, yoga) because as an embodied human her body takes in enormous information and she can feel when things are right.
• Prompt: invest a little in honing / listening to your intuition — learn to notice, interrogate, and contextualize it (is your body tense because of your kid or an impending outage?). No real downside — worst case, a mildly interesting waste of time. (Maybe learn tarot.)

8.2 Embodiment

• Spend more time being in your body instead of living through GPUs — experience your body’s natural rhythm to navigate the pace mismatch. Even a couple minutes of “how do I feel in my body?” has no downside and may just help you feel better.

8.3 Questioning perpetual growth

• This is a very smart room — is it within our locus of control to affect change in the industry? Maybe; we’ve done it before.
• The social order is predicated on continuous growth — when is there enough? There’s no answer (“we have to mine in space next”). Every technological wave promised more leisure (the internet, the microwave marketing) — instead life gets denser (microwaving while vacuuming while running the dishwasher). We create no leisure time.
• Challenge: how can the people in this room (not Bezos/Musk) capture AI’s value for more leisure — enjoying kids, hobbies, weather? “What if this amount of growth is just enough?” Pursuing money with that vigor is idolized, but pursuing anything else that hard we’d call an addiction. “What grows continuously? Cancer.”

9. Exhausting and Exciting

• It’s exhausting and exciting — “you may have more than one feeling at the same time, even contradictory ones.”
• Grace Hopper slide: just as past inventions (the compiler) could have been shaped for a more equitable industry, now it’s our turn to “shake things up in a brand-new way.” Cat, who “had an awful time coming up through tech” as a woman, sees a chance to make the next year more equitable — “what am I going to do about it?” is everyone’s opportunity.
• Deep gratitude for the audience’s time in the “sweaty tent with weird philosophy girl.”

10. Q&A

Q1 — Practices to resist AI-driven acceleration without becoming less competitive?

• Prioritize — ask whether efforts actually move the needle, and define what that means.
• Anchor in things that are real (e.g., regulator timelines that don’t accelerate) rather than the felt contraction of the present; explicitly decide where you apply effort (not everywhere); a material design is not automatically a commitment.
• Leave the building — go outside, museum, bike ride, coffee (preferably together): research links it to more creative outcomes and it energizes people.

Q2 — Is AI fundamentally changing what good engineering craftsmanship means?

• Yes, of course. Mixed feelings: she’d love the end of middle-of-the-night pages (“disk space critical”). AI may let us focus on bigger things; unsure what happens closer to the metal (threading models, assembly like MACRO-32) — you don’t need assembly to be a great programmer, and she’s excited about a hardware-engineering resurgence.

Q3 — One organizational thing/hobby that became actively harmful in the AI era?

• Layoffs. As “evil dictator of the world” she’d ban layoffs and instead reduce everyone’s weekly hours — leisure time creates societal value; maybe “unions 2.0.”

People, Companies & References Cited

• Cat Swetel — speaker; GM, Nubank foundation/infrastructure unit; background in science & technology policy.
• Nubank — Latin-American bank, ~130M customers; Datomic database (append-only).
• Jen — colleague who can speak to Datomic.
• Hartmut Rosa — sociologist; acceleration theory and “frenetic standstill.”
• Little’s Law — arrival vs. departure rate / work-in-process.
• Grace Hopper — the compiler; inspiration for reshaping the industry.
• Cultural references: fast food, the microwave and internet “leisure” promises, GitHub/AWS incidents, egg-freezing perks.
• Concepts: pace mismatch, premature materialization, shrinking present / nested strategy horizons, embodiment & intuition, questioning perpetual growth.

Video: https://www.youtube.com/watch?v=NLbT-DkW5_U — Transcript via yt-transcript.sh; exhaustive outline generated from the full transcript.