IBM asked what humans do when AI does the work. Ian Beacraft answered with org architecture. Aza Raskin answered with whale songs
Opening Pulse
Between sessions, in the Grand Ballroom corridor, something was already running.
A live data-to-sound installation converting plant bioelectricity and room energy into a real-time soundscape. A flutist threading human breath into the signal. Biology as creative force. Plants eating light, turning it to sound, a musician laying the human layer on top.
// sys.prelude
biology already composing
human breath added to signal
day begins from the rootsSessions
When AI Does the Work, What Do Humans Do Next?
// session.log
Justina Nixon-Saintil // IBM
Amanda Butler // Side by Media
subject = redeployment / entry-level hiring / voice driftBy now, most rooms are asking some version of the same question: what happens to human work as AI absorbs more execution? So we’ll keep this quick.
IBM’s answer was pragmatic: AI across roles, yes. Agents take more task execution, human staff move toward consulting and strategic roles, and the workforce remains stable.
Amanda Butler told her teams: I know you did not learn to use em dashes all over your emails. You don’t talk like this. Keep the communication human.
Key points
IBM is framing AI as redeployment infrastructure more than elimination engine
Entry-level hiring remains a strategic investment
AI fluency is becoming baseline
Voice drift is already one of the clearest cultural artifacts of AI use
How to Design a Company AI Can’t Outpace
// session.log
Ian Beacraft // Signal and Cypher
subject = AIOS / culture layer / execution collapseWhat are we going to do with the time AI gives back?
We hear it again: more work.
Beacraft said it like the decision had already been made.
Most organizations still think of AI as a tool. That framing creates friction. So how do we protect orgs from eventual AI outpacing? Switch the frame from tool to system design, from isolated use to agentic workflow, and the logic changes.
That’s the talk.
Klarna’s mistake was the proof point. Cut seven hundred people, then scramble to hire back. They removed the wrong layer because they misunderstood the system. Coordination and culture were not separable.
When we investigate the structural underlay, we find: the cost of execution has collapsed. Building a prototype is now cheaper than the meeting to plan it. Reality is now the faster decision-maker. Build three. Deploy all. Let the world sort it.
Every company is two things pretending to be one: a coordination system and a culture. AI is absorbing more of the first. The second has no replacement.
That makes culture the new frontier.
His answer was AIOS: Architected Information for Organizational Systems. Give the system context all the way down to values. A constitution, not a prompt.
It knows everything about all data.
It knows nothing about you.
So teach it.
Key points
Tool framing creates friction; system framing changes the operating model
Execution cost has collapsed faster than coordination cost
Klarna is a live warning about misreading the org layer
AIOS treats values as infrastructure
Organizations need to be grown, not built
// Cyborg note: Beacraft is describing the cyborg organization. A living system that routes work through agents and holds culture in the one layer the machine cannot inherit on its own.
Powerful AI, Terrible UX: Designing a Smarter UX
// session.log
Jason Hiebl // Extreme Networks
Sudha Narayanan // Extreme Networks
John Gower // Dialect
subject = enterprise agentic UX / trust / context / cognitive load
status = recorded_only / return_to_sourceI mostly caught this one through recording while moving between expo-floor obligations, but the signal was clear:
Chat is not enough.
If users have to become prompt engineers, the product has failed. That was the core argument.
Empty chatbots create cognitive load. Enterprise AI needs context, role awareness, and predictable behavior. Chat may be the entry point, but it cannot be the whole product. Not when real work, real trust, and real workflow complexity are involved.
Sudha Narayanan pushed for systems that infer intent instead of forcing users to translate themselves into machine language.
John Gower emphasized reducing friction from intent to action and making trust visible through transparency, confidence, and “show your work.”
Control is a requirement.
Human-before-the-loop. Human-on-the-loop. Human-in-the-loop. Human-above-the-loop.
Show the receipts. Show the reasoning. Show the resource cost. Show the outcome. Trust in enterprise AI is not a vibe. It has to be designed, measured, and earned.
Key points
If users have to become prompt engineers, the product has failed
Chat-first UX is insufficient for enterprise systems carrying real cognitive load
Trust depends on control, predictability, and transparency
“Show your work” is becoming a core UX requirement for agentic systems
Context-aware, workflow-native experiences matter more than empty chat interfaces
// Cyborg note: Empty chatboxes are a design tell. They reveal a system that still expects the human to do too much translation work.
Nature Speaks. Can AI Help Us Listen?
// session.log
Aza Raskin // Earth Species Project
subject = living intelligence / animal communication / embedded spaceThe loudest room of the week, and the subject was animals.
Raskin opened with the connective thread beneath humanity’s largest problems: narrow optimization at the expense of the whole organic system. We prioritize local wins with system blindness. Disconnection from ourselves and everything else.
Then he played a sound.
Arctic footage. A squeaking call. Nobody got it right. Bearded seal mating call.
That was the point.
The evidence. Parrot parents whisper names in their babies’ ears before they hatch. The babies keep those names for life. Elephants do this. Whales do this. Crows in Lyon wear data backpacks so next-gen researchers can hear: sixty percent of their vocalizations are whispers between individuals in flight. Whales, he said, appear to spend half their acoustic lives using uniquely identified vocalizations. Names.
Half.
He let that hang and the room held it. I saw tears.
The underlying method is embedded-space mapping. AI doesn’t need to know what a signal means in advance. It learns the relationships between signals. Dog means dog means dog across languages not because of the word, but because of the relational shape the embedded-space map draws.
Earth Species Project is asking whether that same relational logic can map animal communication.
Raskin says yes.
The results are beginning to suggest he’s right.
His heuristic was the sharpest epistemological claim of the week:
If you make the problem space bigger, the problem gets easier.
That runs against legacy reductionism, in science, and in infrastructural development. The old instinct is isolate, narrow, compress. Raskin argues that expansion reveals structure.
Then the room got stranger in the best way.
Dolphins check their abs in mirrors. Chimps check their butts. Chimps also spin themselves dizzy and beat the insides of trees to enter trance states. Whales may carry thirty-four million years of oral tradition.
We have not read it yet.
He closed in a place tech keynotes almost never go. To be a good cyborg, you must connect back to nature as the world becomes more technological.
Key points
Narrow optimization remains the underlying pathology
Embedded-space methodology maps relationships before meaning
Animal communication becomes legible when listened to at scale
“Expand the problem space” reverses standard reductionism
Whale oral tradition is now being treated as a real archive
Listening is becoming a design problem, not just a science problem
// Cyborg note: Raskin is building living intelligence. Webb’s convergence pointed inward toward human augmentation. Raskin points outward toward everything else sharing the system. Same loop. Different direction.
Day Theme
Beacraft says organizations need to be grown, not built. Raskin is already working with coevolutionary biology as method. If users have to become prompt engineers in LLM UX, the system has failed. Empty chatboxes, org constitutions, whale names, agentic workflows, ancient oral traditions. Different domains. Same pressure point.
Understanding.
Not just whether communication is possible. Whether the context is shared, and intent survives translation. Whether a system can carry meaning without forcing the human to do all the interpretive labor by hand.
The week has kept asking what part of the human survives when the machine keeps getting better.
The thematic reframe is to forget about which human capacities survive. Ask what was already here. Thinking. Understanding. Building culture.
Offload the execution, not the thinking. Then go find out what was thinking before you arrived.
// sys.daylog · day_05 · 2026-03-16
execution_cost = collapsed
coordination_cost = unchanged
human_value = intent + judgment + context
org_model = grown_not_built
animal_comms = partially_legible
living_intelligence = active / outward facing
constitution_status = pending
problem_space = expand_not_compressClosing
Raskin’s question carries forward as a design question.
If embedded-space methodology can map meaning across language, image, sound, and bioelectricity, then listening is no longer the hard part.
The harder question is obligation.
What do we owe the things we have been ignoring?
If culture is the new frontier and nature is the oldest database, who is responsible for the translation?