strategyAIquantumtechnology

firms that move first

are always structurally

unmatchable.

fusing the
next medium

design thinking and artificial general intelligence have converged into a single, powerful cognitive substrate.

we design living strategic systems in which human insight and machine intelligence continuously strengthen each other.

most strategy goes stale the moment it leaves the room.

we create dynamic cognitive environments that think alongside leaders.

strategic thought now runs through a nervous system that never switches off.

the horizon

capability is arriving in stages, not as a single event. we label our confidence honestly at every stage.

nowobserved

narrow capability, broad deployment

Today's systems are extraordinarily capable within a defined task: drafting, coding, analysis, classification, conversation. They do not set their own goals, and they do not reliably operate across a long horizon without a human checking in. The capability is real and already economically significant. The autonomy is bounded.

—Individual task productivity rises measurably wherever a task can be specified clearly

—The bottleneck shifts from 'can the model do this' to 'do we trust our process enough to let it'

—Competitive advantage accrues to organizations with the operating discipline to deploy well, not the ones with access to the best model

near termemerging

coordinated, multi-step agency

Systems that can plan a multi-step task, use tools, check their own work, and operate for extended periods with only periodic human checkpoints are already in early production. The frontier is moving from single agents to coordinated teams of specialized agents handling genuinely complex workflows end to end.

—Entire workflows, not just tasks, become candidates for automation, with humans moving to supervisory and exception-handling roles

—Organizational design starts to matter as much as model capability: who is accountable when an agent acts, and how is that action reviewed

—The firms that win this era are the ones that build governance and audit infrastructure before they need it, not after an incident forces the issue

medium termanticipated

cross-domain reasoning at expert level

A system that performs at a credentialed expert's level not in one narrow domain but across the range of domains a senior professional moves between in a single day, legal, financial, technical, communicative, is a meaningfully different kind of tool than what exists today. This is where most serious forecasting disagreement begins.

—The economics of expertise change: judgment that was scarce and expensive becomes available at a different price point and speed

—The professional roles that survive are the ones built around accountability and relationship, not information synthesis

—Organizations that have not already rebuilt their operating model around AI-augmented expertise will find the gap very difficult to close quickly

long horizon, genuinely uncertaincontested

general capability

Systems with the flexible, transferable, self-directed problem-solving capability associated with general intelligence remain, honestly, a contested forecast. Serious researchers disagree meaningfully on both timeline and even on what would constitute sufficient evidence that it has arrived. We are not going to manufacture false certainty here.

—Strategic planning at this horizon is about resilience and optionality, not point predictions

—Governance, safety, and control questions become first-order business questions, not back-office compliance items

—The organizations best positioned are the ones who treated every prior era as practice for institutional adaptation, not the ones who waited for certainty before building capability

what Changes

this cuts across function, not industry. it changes how a company staffs, structures, governs, and competes.

The conversation about AI and jobs is usually framed as a binary: automation replaces a role, or it does not. that framing was already wrong for the current generation of tools and will be more wrong for what is coming. What actually happens is narrower and more disruptive: the specific tasks that made up sixty percent of a role get absorbed, and the remaining forty percent, judgment, accountability, relationship, becomes the entire job description.

Organizations that wait for this to become obvious before acting will spend years re-hiring and re-training under pressure. The ones who move now are rewriting role definitions, career ladders, and hiring criteria around the durable forty percent before the transition forces their hand.

This is not primarily a technology project. It is an organizational design project that happens to be triggered by technology, and it belongs on the same strategic planning cycle as workforce planning, not in an innovation lab side project.

workforce and talent strategy

“The job does not disappear. The job description does, and most organizations have not started rewriting it.”

Management structure has historically been shaped by how many direct reports one person can effectively oversee, typically somewhere between five and ten. That constraint exists because human judgment, coaching, and review take real time per person managed.

Agentic systems change the shape of that constraint, not by removing the need for oversight, but by changing what oversight looks like: reviewing outputs and setting direction at a different cadence than managing a team of people day to day. Organizations built around the old span-of-control assumption will carry structural overhead that newer, AI-native competitors will not.

This does not mean flattening every hierarchy overnight. It means treating organizational structure as a variable to be redesigned deliberately, the same way a company redesigns its structure after a major acquisition, rather than something that simply evolves on its own.

organizational structure and span of control

“When a senior person can supervise ten agents instead of ten people, the org chart that was optimized for human management spans stops making sense.”

As systems move from answering questions to taking actions, autonomously sending communications, executing transactions, modifying records, the question of accountability becomes immediate and concrete rather than theoretical. Who is responsible when an autonomous system makes a consequential error is not a question that can be answered after the fact.

Mature governance for this era looks like permission scoping defined before deployment, complete audit trails for every autonomous action, and clear escalation paths for the cases a system is not confident enough to handle alone. This is closely analogous to financial controls: nobody would let a junior employee move money without an approval chain, and the same discipline applies to an agent with equivalent authority.

Boards and risk committees are increasingly going to expect this kind of structure to already exist, not to be built reactively after the first serious incident. Organizations that treat AI governance as a standing agenda item now will not be improvising under regulatory or reputational pressure later.

governance, risk, and accountability

“An autonomous system acting on a company's behalf is a governance question before it is a technology question, and most governance functions are not yet structured to answer it.”

When every competitor in an industry has access to comparable underlying model capability, the differentiator stops being which model a company uses and becomes how effectively that capability is integrated into proprietary data, existing workflows, and institutional decision-making.

The organizations building durable advantage right now are the ones treating their own operational data and workflow knowledge as the genuinely scarce asset, the thing a competitor cannot simply purchase access to, and building the infrastructure to apply AI capability against that asset specifically.

Speed of institutional learning is becoming a competitive variable in its own right: organizations that can run more deployment cycles, learn from more production feedback, and adapt their approach faster will compound an advantage that is very difficult for a slower-moving competitor to close, even with equivalent technology access.

This reframes a question many boards are asking incorrectly. The question is rarely whether to adopt a given model. It is whether the organization has the data architecture and workflow integration discipline to turn shared, commoditized model access into an advantage that is specific to them. That discipline, not the model subscription, is what shows up on the balance sheet over time.

competitive advantage and moat

“Model access stopped being a durable advantage the moment frontier models became broadly available. The moat is moving to data, workflow integration, and institutional learning speed.”

Hypothesis generation, literature synthesis, simulation, and early-stage design iteration are all tasks where current and near-term systems can meaningfully compress timelines, not by replacing researcher judgment but by removing the grunt work that used to consume the majority of a research cycle.

The organizations capturing this advantage early are restructuring R&D workflows around faster iteration cycles deliberately, rather than treating AI tools as a productivity bonus layered onto an unchanged process. A research process designed for a six-month iteration cycle does not automatically benefit from a tool that could support a six-week one; the process itself has to be redesigned.

There is a genuine risk on the other side worth naming honestly: faster idea generation without a matched increase in rigorous validation capacity just produces more unvalidated ideas, faster. The organizations getting real value here are investing as much in validation throughput as they are in generation throughput.

research, development, and product innovation

“The bottleneck in R&D has historically been generating and testing ideas at scale. That bottleneck is loosening faster than most innovation functions have adjusted their planning cycles for.”

As agentic systems become capable of handling genuinely complex customer interactions end to end, the operational question shifts from 'can a system handle this' to 'what level of autonomy is appropriate for this specific interaction type,' which is a risk and brand question as much as a technical one.

The organizations getting this right are not pursuing blanket automation. They are mapping interaction types by complexity and stakes, and deliberately deciding where full agent autonomy is appropriate, where agent-assisted human interaction is the right model, and where human handling remains non-negotiable regardless of technical capability.

The reputational cost of a poorly governed automated interaction going wrong publicly is now a board-level risk, not a customer service metric. This is why customer experience strategy and AI governance, which are usually owned by entirely separate functions, need a shared decision framework rather than two parallel, uncoordinated efforts.

customer experience and relationship management

“Customers will increasingly be unable to tell, and will increasingly not care, whether a high-quality interaction was handled by a person or a well-governed agent. What they will notice immediately is a poorly governed one.”

what we’d push back on

claims we hear often, and don’t agree with.

“AGI is just a more powerful version of the chatbots we use today.”

↓our view

Today's most capable systems are narrow in an important sense: they perform extraordinarily well within a defined task but do not reliably set their own goals or operate across a long horizon without checkpoints. General capability implies flexible, transferable problem-solving across genuinely novel domains, a meaningfully different kind of system, not simply a scaled-up version of what exists now.

“once AGI arrives, it will be obvious and undeniable to everyone at once.”

↓our view

We think this is one of the more dangerous assumptions in circulation. Capability is very likely to arrive unevenly, domain by domain, with genuine disagreement among serious researchers about whether a given threshold has actually been crossed. Organizations waiting for an unambiguous signal may be waiting for something that never arrives in the clean form they expect, while competitors who acted on the uneven signal already have already built the advantage.

“if AGI is still years away, there is no urgency to act now.”

↓our view

This treats AGI as the only relevant threshold, when the economically significant disruption is happening at every stage along the horizon, not just at the final one. Workforce strategy, governance infrastructure, and data architecture decisions made now compound in value regardless of exactly when, or whether, full general capability arrives on the timeline any particular forecaster expects.

“the safest strategy is to wait and see what competitors do first.”

↓our view

This works only if the capability being waited on arrives as a discrete event a company can react to quickly. Under the more likely uneven-arrival scenario, the advantage accrues to whoever has spent the waiting period building institutional capacity, data infrastructure, and governance, not whoever reacts fastest after the fact. By the time the signal is unambiguous enough for a 'wait and see' strategy to act on, the early movers have already compounded several iteration cycles of advantage.

starting now

none of this requires waiting for certainty. every item below is worth doing regardless of exactly when the next threshold is crossed.

audit task composition before you redesign roles

Before rewriting a single job description, map what the role actually does at the task level. Most organizations are working from job titles, not task inventories, and the redesign work is impossible to do well without that detail.

build the governance structure before you need it

Permission scoping, audit trails, and escalation paths for autonomous systems take real time to design properly. Building this after an incident forces the question is materially worse than building it as standing infrastructure now.

treat your own data as the moat, not the model

Model access will keep commoditizing. The lasting advantage is in how well a system is integrated against an organization's own proprietary data and workflow knowledge, which only that organization can build.

run real deployment cycles, not pilots that never graduate

Institutional learning speed is a genuine competitive variable. Organizations that ship, measure, and iterate in production learn faster than organizations that run an indefinite pilot phase waiting for certainty that will not arrive on its own schedule.

see the technology

the capability index →

questions we get

on the record about what we don’t know too.

We use the term carefully, to mean systems with flexible, transferable problem-solving capability across domains, comparable to a skilled human professional moving between unfamiliar problems, rather than as a marketing label for any sufficiently capable current model. We are explicit on this page about which claims are observed today, which are reasonably near-term, and which remain genuinely contested among serious researchers.

No, not by the definition above, and we think claims that it has already arrived are usually either marketing or a much looser definition than the one most researchers use. What has arrived, and what is economically significant today, is narrow capability deployed broadly, plus early, genuinely useful agentic coordination. The cross-domain, self-directed capability associated with general intelligence is still ahead, on a timeline serious people reasonably disagree about.

By building institutional capacity rather than betting on a specific date. Governance structure, workforce model redesign, and data infrastructure are valuable regardless of exactly when the next capability threshold is crossed, which makes them a sound investment under genuine uncertainty rather than a bet that can be wrong.

Treating it as binary: either dismissing the trajectory entirely because today's systems still have clear limits, or treating a speculative future capability as already deployable and making decisions accordingly. Both lead to poor outcomes. The more useful posture is building the operating discipline to absorb capability as it actually arrives, function by function.

We deliberately do not publish a single fixed date, and we are skeptical of anyone who states one with high confidence. What we offer instead is a horizon view, openly labeled by how confident the evidence is at each stage, because we think that is more useful and more honest than false precision.

human insight

machine intelligence

sustainable competitive advantage.

Creating compounding effects that continuously strengthen organization's ability to sense, respond, and adapt to change.

design for AGI

the next era starts here →

fusing the
next medium

design thinking and artificial general intelligence have converged into a single, powerful cognitive substrate.

we design living strategic systems in which human insight and machine intelligence continuously strengthen each other.

most strategy goes stale the moment it leaves the room.

we create dynamic cognitive environments that think alongside leaders.

strategic thought now runs through a nervous system that never switches off.

the horizon

capability is arriving in stages, not as a single event. we label our confidence honestly at every stage.

nowobserved

narrow capability, broad deployment

—Individual task productivity rises measurably wherever a task can be specified clearly

—The bottleneck shifts from 'can the model do this' to 'do we trust our process enough to let it'

—Competitive advantage accrues to organizations with the operating discipline to deploy well, not the ones with access to the best model

near termemerging

coordinated, multi-step agency

—Entire workflows, not just tasks, become candidates for automation, with humans moving to supervisory and exception-handling roles

—Organizational design starts to matter as much as model capability: who is accountable when an agent acts, and how is that action reviewed

—The firms that win this era are the ones that build governance and audit infrastructure before they need it, not after an incident forces the issue

medium termanticipated

cross-domain reasoning at expert level

—The economics of expertise change: judgment that was scarce and expensive becomes available at a different price point and speed

—The professional roles that survive are the ones built around accountability and relationship, not information synthesis

—Organizations that have not already rebuilt their operating model around AI-augmented expertise will find the gap very difficult to close quickly

long horizon, genuinely uncertaincontested

general capability

—Strategic planning at this horizon is about resilience and optionality, not point predictions

—Governance, safety, and control questions become first-order business questions, not back-office compliance items

—The organizations best positioned are the ones who treated every prior era as practice for institutional adaptation, not the ones who waited for certainty before building capability

what Changes

this cuts across function, not industry. it changes how a company staffs, structures, governs, and competes.

workforce and talent strategy

“The job does not disappear. The job description does, and most organizations have not started rewriting it.”

organizational structure and span of control

“When a senior person can supervise ten agents instead of ten people, the org chart that was optimized for human management spans stops making sense.”

governance, risk, and accountability

“An autonomous system acting on a company's behalf is a governance question before it is a technology question, and most governance functions are not yet structured to answer it.”

competitive advantage and moat

“Model access stopped being a durable advantage the moment frontier models became broadly available. The moat is moving to data, workflow integration, and institutional learning speed.”

research, development, and product innovation

“The bottleneck in R&D has historically been generating and testing ideas at scale. That bottleneck is loosening faster than most innovation functions have adjusted their planning cycles for.”

customer experience and relationship management

what we’d push back on

claims we hear often, and don’t agree with.

“AGI is just a more powerful version of the chatbots we use today.”

↓our view

“once AGI arrives, it will be obvious and undeniable to everyone at once.”

↓our view

“if AGI is still years away, there is no urgency to act now.”

↓our view

“the safest strategy is to wait and see what competitors do first.”

↓our view

starting now

none of this requires waiting for certainty. every item below is worth doing regardless of exactly when the next threshold is crossed.

audit task composition before you redesign roles

build the governance structure before you need it

treat your own data as the moat, not the model

run real deployment cycles, not pilots that never graduate

see the technology

the capability index →

questions we get

on the record about what we don’t know too.

human insight

machine intelligence

sustainable competitive advantage.

Creating compounding effects that continuously strengthen organization's ability to sense, respond, and adapt to change.

design for AGI

the next era starts here →

fusing thenext medium

fusing thenext medium

fusing the
next medium

fusing the
next medium