For fifty years, organizations optimized for information. The next fifty will belong to the organizations that optimize for intelligence.
We built databases, dashboards, KPIs, OKRs, scorecards, and quarterly reviews to solve the defining problem of the Information Age: how to collect, organize, and move information through a large organization at scale. Those systems worked. They were among the great managerial innovations of the last century.
But the problem they solved is gone. Artificial intelligence has made information abundant and nearly free; the world now creates and replicates roughly 181 zettabytes of data a year, and most enterprises already leave roughly two-thirds of the data they have untouched.[1][2] Information is no longer the bottleneck. Understanding is. And the market’s reflex, pour ever more data into ever bigger models, is the same mistake, now automated.
We are information-rich and intelligence-poor.
This paper makes one argument: enterprises have spent decades building systems of record and almost nothing building systems of context, and in an agentic world, that gap becomes the single most expensive thing in the business. A system of record stores what happened. A system of context stores why, and makes that why durable, connected, and usable by humans and machines alike. For fifty years this was a tolerable weakness, because humans could supply the missing why in real time, in a room. It stops being tolerable the instant agents act at machine speed: they do not pause to ask what a number meant or why a target was set, which makes context (complete, clear, and connected) the control layer that decides whether they create value or compound error. The winners of the next decade will not be the companies with the best dashboards or the biggest AI models. They will be the companies with the richest, most usable organizational understanding. Context is the new moat.
What follows is the proof that the crisis is real, a concrete standard for fixing it, and the reason most organizations will not act until machine-speed execution makes the cost impossible to ignore.
Consider two facts. Placed side by side, they indict half a century of management instinct.
First, information has never been more abundant. IDC estimates the world will create and replicate roughly 181 zettabytes of data in 2025, up from about 6.5 zettabytes in 2012: a stack that has grown more than twenty-seven-fold in little over a decade and continues to compound.[1] By any historical measure, the scarcity that justified half a century of information systems is over.
Second, almost none of it is used. A landmark Seagate and IDC survey of 1,500 enterprise leaders found that 68% of the data available to businesses goes unleveraged; only about a third is ever put to work.[2] We are not short on information. We are drowning in it, and starving for the ability to act on it.
Over two decades in boardrooms, risk committees, and executive strategy reviews, I watched the same gap form in organization after organization, long before the data named it: leaders surrounded by more numbers than any executive in history, and no better able to explain their own results. The dashboards started the conversation; context determined the decision. The instinct, always, was to add another report. It never once closed the gap, because the constraint was never the supply of information. It was the absence of the thing that turns information into a decision. That thing is context.
This is the paradox of the modern enterprise. We solved seeing and mistook it for understanding. For decades that confusion was survivable, because a human could always close the gap in the moment. We are about to lose that luxury: as agents take over execution and run at the speed of software, the distance between what an organization records and what it actually understands stops being a quiet tax and becomes the thing that decides whether those agents help or harm. The rest of this paper is about that gap: why it is suddenly the most important thing on the management agenda, and what to do about it.
Management is not timeless. Each technological era produces the management system its constraints demand, and then mistakes that system for the natural order of things.
The Industrial Age created the factory. The factory created scale. Scale created the need for coordination, and so it created professional management: the org chart, the span of control, the chain of command. These were not arbitrary. They were the operating system for a world whose binding constraint was the coordination of physical work.
The Information Age created software. Software created data. Data created the dashboard, and the dashboard created the KPI, and the KPI created the OKR. Entire industries (business intelligence, analytics, reporting, performance management) grew up around a single, sensible mission: collect information, organize it, and move it to the people who needed it to decide. For most of business history the executive’s problem was that they could not see; they were flying a large, fast aircraft with almost no instruments. The dashboard gave them instruments. It is one of the most useful inventions in the history of management, and the temptation of every manifesto is to burn down the thing that worked. I won’t.
The dashboard was built for a world where information was scarce.
That is the whole point. In a world of scarcity, the highest-value move is to gather and distribute the scarce thing. But the binding constraint has moved. Information is now the most abundant resource in the enterprise, and a management system perfectly engineered for scarcity becomes, in an age of abundance, a machine for producing more of what we already have too much of. Every era’s management system eventually becomes the next era’s drag, and most enterprises are now running one built for a problem that no longer exists.
Let me be blunt about the villain of this story, because if I am vague about it the whole argument goes soft. The villain is not the dashboard. The dashboard is a tool, and a good one. The villain is a belief so widely held that we have stopped noticing it is a belief at all:
“More information creates better decisions.”
It sounds obviously true. It is the unexamined assumption underneath nearly every enterprise software purchase of the last thirty years. Add another data source. Add another report. Add another real-time feed. Surely a more-informed organization is a better-deciding one.
It is not. Past a fairly low threshold, more information does not produce better decisions. It produces more confident decisions, which is a different and more dangerous thing. It produces meetings that feel rigorous because there are a lot of numbers on the screen. Decisions are not bottlenecked by the availability of facts; they are bottlenecked by understanding: by knowing which facts matter, how they connect, what they imply, and what was true the last three times something like this happened. None of that arrives with the fortieth widget. We have been treating a context problem as an information problem, buying more of the former to fix the latter, and wondering why judgment never improves.
This is worth stating carefully, because the rest of this paper calls for capturing more, and I am not contradicting myself. I am not arguing for more data. I am arguing for more of an entirely different and far scarcer thing: the recorded why behind the numbers, captured with clarity and discipline. High signal, not higher volume. The dashboard answered the scarcity problem. It was never the answer to the judgment problem, and AI is about to make the difference impossible to ignore.
Here is the sentence at the center of this paper: enterprises have spent decades building systems of record, and almost no time building systems of context.
A system of record stores what happened. Your ERP is a system of record. So are your CRM, your data warehouse, your BI stack, your financial close, your risk register, and your OKR tracker. They are excellent at the verbs they were built for (capture, store, query, display) and ask any of them what happened and you will get a fast, accurate answer.
A system of context would store why it happened, and make that why durable, searchable, and transferable. Why did we set this target where we did? What did we assume? What did we trade away, and who decided? What did we try before, and how did it go? When the number moved, what was the actual cause, as understood by the people closest to it? Almost no organization has a system for that. The context exists; it is simply stored in the worst possible medium: human memory, distributed across people who are busy, who change teams, who leave, and who were never asked to write it down. We have built a civilization of perfect recall for facts and near-total amnesia for reasons.
The costs are everywhere once you can see them. When a key person leaves, the context leaves with them; the record is intact and the understanding is gone. When teams reorganize, context fragments, and the connection between the half of the story sales held and the half finance held quietly disappears. When companies scale, context dilutes, because the founders carried the why in their heads and there was never a system to propagate it. And when the same decision comes round again, it gets re-litigated from scratch, because the record of the last attempt is a number in a database with no story attached.
Every metric has a story. We built systems that keep the metric and throw away the story.
This forgetting is not free, and it is not rare. Knowledge workers already lose roughly 9.3 hours a week, close to a full day in five, simply hunting for information that exists somewhere but cannot be found, or for the colleague who remembers the context behind it.[3] Poor data quality, much of it a context and definition problem, costs the average organization an estimated $12.9 million a year.[4] These are the visible edges of an invisible crisis, and they were measured before AI put the organization’s own reasoning on the critical path. Agents do not reduce the cost of forgetting; they multiply it.
The most expensive version of this is institutional. When NASA wound down Apollo, it did not just retire hardware; it dispersed the workforce, dismantled the tooling, and lost the tacit engineering knowledge of how the real vehicles differed from the drawings. The blueprints largely survived. The understanding did not, which is why rebuilding a Saturn V from scratch would be extraordinarily costly and slow even today, decades later.[5] The lesson is not about rockets. It is that a system of record is not a system of context, and an organization that confuses the two is one reorg away from forgetting how it works.
This is the Context Crisis, and it has hidden in plain sight precisely because it never appears on a dashboard; by definition, it is the stuff the dashboard cannot hold. For most of the Information Age the gap was tolerable, because humans could close it in real time, in a room. AI is about to make that gap the most expensive thing in the enterprise. To see why, look at the ladder we keep stopping halfway up.
Picture a ladder. Each rung answers a different question, and each depends entirely on the one beneath it.
| Layer | The question it answers | Where most organizations live |
|---|
| Data | What exists? | Stored everywhere |
| Information | What happened? | Most dashboards stop here |
| Context | Why did it happen? | Lives in people’s heads |
| Intelligence | What should we do next? | Rare and mostly tacit |
| Wisdom | How should we adapt over time? | Almost never institutional |
Most organizations live on the second rung. They know what happened and stall there. The leap they never make is to context, why it happened, which is where facts become decisions, and which every rung above depends on. You cannot reach intelligence by piling up more information; you do not reach the top of a ladder by widening the bottom rung. Making that climb something a system can do, not just a veteran in a room, is the work of the next era, and the rest of this paper is about how.
In 2025, MIT’s NANDA initiative studied enterprise AI and put a hard number on it: roughly 95% of generative-AI pilots were delivering no measurable impact on the P&L. The headline got the attention; the diagnosis deserved more of it. The gap was not explained by model quality. It was a learning gap: the tools, and the organizations deploying them, lacked the context to apply intelligence to the specific situation in front of them.[6]
Say it plainly, because the entire AI conversation is pointed in the wrong direction:
An AI system without context is a very sophisticated guesser.
Give a capable model your dashboard and it will describe what happened fluently. Ask it why, and it will produce a confident, plausible answer that is too often wrong, because the actual reason was never recorded anywhere it could read. The reason lived in the head of the regional VP who remembered the supplier you dropped, the commitment you made, the thing you already tried. The model was not dumb. It was uninformed in exactly the way the organization is uninformed about itself; it just hides it better.
This is why a veteran executive still routinely outperforms an expensive AI deployment. It is not that the human processes data faster; the machine wins that contest easily. It is that the human carries the context, the why behind the numbers, and the machine has been handed only the what. We are pouring intelligence into organizations that have no system for the one input intelligence most depends on, and then expressing surprise when the results disappoint.
The companies that win the AI era will not be the ones with the largest models. Models are converging and commoditizing; your competitor can rent the same one you can. The durable advantage is the proprietary thing the model is grounded in: the structured record of how your specific organization thinks, decides, learns, and connects. That asset cannot be downloaded, copied, or bought off the shelf. It has to be accumulated.
Context is the new moat. It is the one input to AI your competitors cannot acquire off the shelf.
Most AI initiatives will disappoint not because the technology fails, but because it is being layered on top of the Context Crisis instead of being used to solve it. Automating a process you do not understand simply lets you be wrong faster and at greater scale. The organizations that treat AI as a context-building engine, capturing the why, connecting it, feeding it back, will pull away from the ones that treat it as a faster dashboard.
The market has already diagnosed the problem and prescribed the wrong cure. Watch what the largest software companies are doing. Salesforce, Slack, Google, Microsoft: nearly every major system of record is racing to ingest as much data as it can hold and to promise that piping all of it into a model will finally give agents the context they need to make aligned decisions and take helpful action. It is a seductive pitch. It is also the dashboard era’s founding belief, more information creates better decisions, now with a model bolted on.
Hoarding more data is not the same as supplying the right context.
An agent handed everything is not closer to the answer; it is further from it. Context is not volume. It is the why behind the number, the target the number is judged against, the risks it connects to, the owner accountable for it, and the post-mortem from the last time it moved. None of that lives in the raw exhaust of a CRM, a data lake, or a chat history, no matter how much of it you pump into a model. You can ingest every message your company ever sent and still not know why the board approved the project, what you assumed when you set the target, or what you already tried and abandoned. The incumbents are scaling the haystack and calling it context. That gap is the entire opportunity, and the reason the advantage will not accrue to whoever holds the most data, but to whoever holds the most usable understanding.
And the bridge to making any of this pay is not a bigger model or a bigger data lake. It is context that matters: the structured why that finally lets an agent make sense of the unstructured data already sitting in your systems of record. Done right, the right context does not replace those systems; it makes them usable, turning dormant data into grounded action.
This is the line the rest of this paper walks: not more data, but the right context; not activity, but outcomes; not a faster dashboard, but a system that understands.
If the gap between record and context has existed for fifty years, a fair question is: why does it suddenly matter? Why 2026, and not 2015, or 2035?
The answer is that, for the first time in history, context has become economically valuable. For all of business history the why lived in human heads, and that was a defensible place to leave it, because only humans could use it. No machine could read an organization’s reasoning, connect it, and act on it, so under-investing in capturing context was, in cold economic terms, rational. You could not earn a return on it.
For the first time, context pays. That is what changed.
Three capabilities arrived at once. AI can now consume context, reading and reasoning over the why behind your numbers at a scale no human team ever could. It can generate context, helping capture, structure, and keep that why current instead of letting it evaporate. And it can act on context, executing through agents, at machine speed, on the understanding you give it. The moment software can consume, generate, and act on context, the why stops being organizational hygiene and becomes the highest-leverage asset a company can build, and the most expensive one to lack.
That is why the timing is specific. In 2015, the models could neither consume nor act on context, so building it earned no near-term return. By 2035, the leaders who start now will have compounded a decade of organizational understanding that latecomers cannot buy or copy. 2026 is the narrow window where the capability to use context has finally arrived and the discipline to build it has not. That gap, between what AI can now do with context and how little of it enterprises have bothered to capture, is the whole opportunity. And like every such gap, it closes fast.
There is a deeper reason the timing matters, and it has nothing to do with hype. Two curves are crossing. The cost of capable AI is collapsing while the cost of human labor is not, and once that gap opens, capitalism does the rest.
The numbers are stark. The inference cost of GPT-3.5-level performance fell roughly 280-fold in two years, from about $20 to $0.07 per million tokens, while at the hardware level costs fall around 30% a year and energy efficiency improves about 40%.[7] Capability is climbing as fast as cost is falling. When a unit of capable digital work gets an order of magnitude cheaper every year and its human equivalent does not, substitution stops being a matter of taste. It is arithmetic, and the pressure it creates points in one direction: toward the most agentic operating model an organization can run.
This isn’t ideology. It’s arithmetic, and arithmetic does not negotiate.
The serious forecasts argue only about speed. Anthropic’s Dario Amodei warned in 2025 that AI could eliminate up to half of entry-level white-collar jobs and push unemployment to 10–20% within one to five years, concentrated in finance, law, consulting, and technology.[8] Goldman Sachs estimates that roughly 300 million jobs worldwide are exposed to automation, with about a quarter of all work hours in advanced economies automatable.[9] You can argue the transition takes three years or ten. It is far harder to argue it will not happen. The destination is not really in dispute; only the timetable is.
And the timetable is the one variable you only partly control, because for many, the market will set it for you. A SaaS company whose competitors ship and operate at agent speed cannot choose to move slowly; it adapts or it is repriced. A regulated incumbent has more room, but not infinite room. Whether the shift arrives as an opportunity or as an existential squeeze, it arrives.
Right now the market is loud with the opposite story: AI is too expensive, the pilots do not pay back, the spend dwarfs the return. Goldman Sachs’ own head of equity research called it “too much spend, too little benefit,” doubting that a trillion dollars of AI investment would ever earn out, and questioning whether costs would fall at all.[10] But that case rests on a premise the data is already demolishing: that AI stays expensive. It does not. The very two years that produced the skepticism also produced a 280-fold collapse in the cost of intelligence. Today’s price objection is a snapshot mistaken for a trend. Models will only get cheaper and more capable, and every step down that curve tightens the math.
There is also a reason the first wave felt costly and underwhelming. It was measured by activity (tokens consumed, prompts generated, seats sold, models deployed), the very metrics the AI vendors put forward, because those are the metrics they bill against. Consumption is a wonderful business model for a model provider and a terrible scorecard for a buyer. The next wave will be measured by outcomes: decisions improved, risks reduced, revenue created, costs avoided. That shift, from activity to outcomes, is the whole game, and outcomes are exactly what context makes possible. The transition will not be simple; it demands the unglamorous discipline of capturing context and the cultural will to change how work gets done. But for the organizations that do the work, the same math that threatens the laggards becomes the engine of the leaders, and the reward is ample.
If this were merely a critique it would be easy to dismiss. It is not. The organizations that perform most reliably under pressure have already built systems of context, deliberately, and at real cost, and they have done so for decades. They are the proof that this is achievable, and a preview of what every enterprise will need.
Toyota: institutionalizing the “why.”
The Toyota Production System is admired for efficiency, but its deeper innovation is a discipline for capturing reasoning. The “Five Whys” forces teams past the symptom to the root cause, and the one-page A3 report makes the why behind every problem and countermeasure explicit, portable, and reviewable. Toyota did not just record what broke; it built a culture in which writing down why, clearly enough for someone else to act on, is part of the job, not an afterthought.[11] The result is an organization that compounds learning instead of repeating mistakes.
Amazon: the Correction of Errors.
Amazon runs a standardized post-incident mechanism called the Correction of Errors, or COE: a written analysis of what failed, the impact, the timeline, the root cause (often via the Five Whys), and the corrective actions, shared so the whole organization learns from a single failure.[12] Paired with its six-page narrative memos, which force argument and assumption onto the page rather than into a slide, Amazon has effectively built a system of context for both its failures and its decisions.
Google: the blameless postmortem.
Google’s Site Reliability Engineering practice is built on the blameless postmortem: after every significant incident, teams document what happened and why, focused on the systemic reasons people had incomplete or incorrect information rather than on assigning blame. The postmortems are written, widely shared, and treated as assets.[13] Notably, Google credits the origin of this discipline to aviation and healthcare, industries where the failure to learn from the last error is measured in lives.
The U.S. Army: After-Action Reviews.
The Army institutionalized reflection at the group level through the After-Action Review, a structured discussion after every significant event asking what was supposed to happen, what actually happened, and why, and then built the Center for Army Lessons Learned to collect, analyze, and disseminate that context across the entire force.[14] It is, in effect, an organizational memory engineered on purpose, so that hard lessons learned by one unit do not have to be relearned in blood by the next.
Systems of context are not theoretical. The most reliable organizations on earth have run them for decades.
Two things stand out. First, every one of these is as much a cultural achievement as a technical one; each required leadership to insist that capturing the why was real work, worthy of senior attention. Second, all of them predate modern AI and were expensive to run by hand. What changes now is that AI can dramatically lower the cost of building and using context, but only for organizations disciplined enough to feed it. The opportunity is to give every enterprise what only the most elite institutions have had, and to make it the default rather than the exception. A caution, to keep this honest: context-keeping is necessary but not sufficient. Each of these institutions paired it with a culture that valued the truth it surfaced, and software can lower the cost of the discipline but never supply the will.
If context is the asset, vague encouragement to “capture more” is useless. An asset needs a standard: a concrete, repeatable definition of what complete context looks like, applied consistently. So here is one, expressed at the level where work actually happens: the individual metric.
The premise is simple. Every metric tied to an objective or a risk should travel with its full story attached. Not a number floating on a dashboard, but a number wrapped in the context that makes it actionable by a human or an agent. A metric that meets this standard is no longer a data point; it is a unit of organizational intelligence.
The first six elements are the operating core: description, targets, owner, status, post-mortem, action plan. The final four are what make context complete in an agentic world: the connections between metrics, the outcomes of actions, the machine-readable clarity that prevents an AI orchestrator from misinterpreting the data, and the history that shows how understanding changed. Most organizations capture one or two of these, inconsistently. None of it is exotic. It is, frankly, unglamorous. That is precisely why almost no one does it, and precisely why doing it is an advantage.
The fair challenge to everything above is that it sounds like something you already bought. You have business intelligence. You have an OKR tracker and a risk register. You may have a knowledge base, a data catalog, and a copilot bolted on top. If context mattered this much, surely one of them already captures it. They do not, and the reasons they do not are worth being precise about, because they are also the reasons this is a new category and not a feature.
Business intelligence shows what happened, beautifully. It is the system of record made visible. It was never built to hold why a target was set, what was assumed, or what connects to what; ask a dashboard “why,” and it has nothing to say. Context is the layer BI leaves empty.
Knowledge management stored documents for humans to search. Context stores reasoning, attached to live objectives, metrics, and risks, and structured for a machine to consume. A wiki full of last year’s decks is not a system of context; it is a graveyard you have to exhume by hand.
Data governance and catalogs make your data clean, compliant, and findable. That is about the data. Context is about the judgment: why this number matters, what it connects to, and what to do when it moves. Perfectly governed data with no context is still mute.
OKR and GRC tools record the targets and the statuses, the what and the how-much. They rarely capture, as first-class and machine-readable, the why behind a target, the post-mortem when it is missed, or the web of dependencies between objectives and risks. They are systems of record for intent, not systems of context.
And a copilot is a consumer of context, not a substitute for it. A retrieval-augmented assistant can only surface what was written down; if the reasoning was never captured, there is nothing to retrieve, and the model fills the void with a confident guess. A copilot pointed at your exhaust is the haystack problem with a chat interface.
So will the incumbents simply add this? It is the right question, and the answer is structural. The giants’ advantage and business model pull the same way: ingest more of your data into their platform. A system of context pulls the other way: it captures your organization’s proprietary reasoning and is most valuable when you own it, in your environment, as your moat rather than theirs. That is an awkward thing to sell for a company whose economics depend on holding your data. Foundation models only sharpen the point: as they grow cheaper and better, the scarce input is no longer intelligence but the grounded context that makes intelligence yours. Better models make context more valuable, not less.
Every tool you already own is either a source the context layer draws from, or a consumer it feeds.
Each is real and useful. None of them is the connective tissue between strategy, performance, and risk: the structured why that makes the others trustworthy. That layer is the system of context, and it is precisely the one no one has built.
Naming the standard, and ruling out the tools that won’t deliver it, is the easy part. The hard part is admitting what it asks for: a renewed, organization-wide commitment to the unglamorous work of writing things down, clearly, in a central place, as a matter of routine. I want to be direct about this, because it is the part most leaders will instinctively resist.
The future will be decided by a discipline that looks, today, like administrative overhead.
Here is the reframe that matters: this is not administrative work, and the moment leadership treats it as administrative, it dies. We have run this experiment before. In the 1990s, companies poured money into knowledge management and even created a Chief Knowledge Officer role to own it. By the 2010s the title had largely faded, not because knowledge stopped mattering, but because the work was treated as overhead, its ROI was hard to prove, and it was delegated away from the people who actually held the knowledge.[15] The lesson is not that context-keeping failed. It is that context-keeping cannot be outsourced to a custodian while everyone else carries on as before.
So when the question comes up (should a company appoint a “librarian,” a single context custodian to own all of this?), my answer is no, and the history above is why. A librarian curates other people’s knowledge, but the context that matters is authored by the people who hold it: the metric owner, the decision-maker, the operator closest to the cause. Centralize it in one role and everyone else concludes it is someone else’s job, and the tacit understanding you needed most is exactly what never gets written down. The proven pattern is the opposite. Data governance learned this the hard way and settled on a federated model: distributed data owners and stewards accountable for their own domains, coordinated by a light central function, with leadership owning the standard. Context should work the same way. Ownership lives at the metric level, which is exactly what the Complete Context Standard requires, and leadership owns the discipline, the cadence, and the consequences of letting it slip.
The right analogy is not the library. It is the financial close. No competent leader calls the monthly close “administrative,” even though it is, mechanically, the disciplined recording of things that already happened. We treat it as sacred because capital markets demanded a rigorous, audited, repeatable record of financial facts, and we built an entire profession and cadence around it. Agentic execution will demand the equivalent for decision context: a rigorous, repeatable record of the why behind the numbers. Call it the context close. The organizations that build that muscle will run circles around the ones still treating it as paperwork.
And there is a new reason this becomes non-negotiable, one the knowledge-management era never faced: machine speed. When execution happened at human speed, ambiguous context was survivable, because a person could pause and ask a colleague what something meant. AI orchestrators do not pause. They act on what is written, at scale, in seconds. Context that is incomplete, stale, or ambiguous does not slow them down; it sends them confidently in the wrong direction, fast. Context stops being organizational hygiene and becomes the control layer for agentic execution. In a world where agents execute down in the layers where work actually gets done, full, accurate, clearly written context becomes the difference between an organization that compounds advantage and one that automates its own confusion. The quality of what you write down will determine the quality of what your agents do. Knowledge management helped humans remember; this determines whether your machines act correctly, which is exactly why it cannot be filed under the same “overhead” that killed the last attempt.
This is why the deepest version of the argument is not about software at all. It is about leadership culture. The commitment to record context, completely, accurately, and with clarity, is a cultural choice that only leadership can make and model. It is, I believe, about to become one of the highest-leverage things a leadership team does. Do I think the future demands this? I do. I think the boring work of recording the why, in a central place, with discipline, is the work that determines whether humans and agents in the execution layers create value or chaos. And I think the leaders who understand that first will quietly build an advantage that their competitors, busy buying more dashboards, will not see coming.
The management model of the Information Age is a pipeline. It runs in a straight line: strategy, then planning, then execution, then measurement, then reporting, then a review meeting, and then, mostly, the line ends. The learning from one cycle rarely makes it back to the start of the next in any structured form. The pipeline moves information forward; it does not compound understanding over time.
Reality is not linear. Reality is adaptive. The operating model the next era demands is a flywheel, a loop that gets heavier and faster with every turn:
Leadership intent → execution → outcomes → signals → context → intelligence → better decisions → refined intent, and around again.
The difference that matters is the context step, the one the pipeline drops on the floor. In a flywheel, every cycle captures not just what the outcome was but why, and that why becomes an input to the next turn rather than evaporating into someone’s memory. Learning stops being an event that happens to lucky teams and becomes a property of the system. Intelligence compounds. Decision quality climbs, not because anyone got smarter, but because the organization stopped forgetting.
This is the architecture underneath RWRD’s own framework, and it is why we describe the loop as define, measure, discover, act, repeat. Define what matters (objectives, targets, risk thresholds) because the human layer always comes first; a flywheel with no one steering just spins. Measure honestly against it. Discover the connections between objectives, performance, and risk that no one can hold in their head at once. Act on the intelligence that surfaces. Then repeat, with the why from this turn carried into the next. The point of the loop is not motion. The point is that each turn leaves the organization understanding itself a little better than the last.
There is one category of context so important, and so badly served by systems of record, that it deserves its own section: culture. The culture strategist Tracey Keele of Keele Advisory puts it more sharply than I can: culture, she says, is “how work gets done” and “the worst behaviors we tolerate.”[16] It is not a soft topic. It is the highest-leverage form of organizational context there is, and it is almost entirely missing from our systems.
Technology amplifies behavior; it does not correct it. Drop powerful tools into a healthy organization and it gets faster and sharper. Drop the same tools into one riddled with ambiguity, turf wars, and unspoken dysfunction and you get chaos at machine speed. AI is the most powerful amplifier we have ever built, which means in the age of agents, culture matters more, not less.
The evidence is not subtle. McKinsey finds roughly 70% of large transformations fail, and that culture, not technology, is the most common cause; its 2026 research found three-quarters of organizations struggling to build the high-performance cultures their strategies assume.[17] These failures do not appear as a line item. They show up as the gap between the plan on the slide and the behavior in the building, which is to say, they are a context problem wearing a different hat.
History makes the stakes concrete. Boeing’s 737 MAX was not, at root, a software failure; multiple investigations described a safety culture eroded by years of cost and schedule pressure and a “culture of concealment” in which engineers felt unable to surface what they knew.[18] The context that could have prevented disaster existed inside the organization; it simply had nowhere safe to go. A generation earlier, Enron and its auditor Arthur Andersen showed the same physics in the opposite domain: when culture rewards the wrong behaviors, no system of record will save you, and a nearly ninety-year-old firm can vanish in months.
Keele names the specific culprits, which she calls Culture Killers: ambiguity, bureaucracy, tribalism, toxic positivity, inconsistency, incivility, complacency, and short-termism.[16] What is striking is that each one is, in effect, a context-destroying machine. Ambiguity guarantees the why is never clear enough to record. Tribalism keeps the two halves of the story in separate buildings. Short-termism optimizes for this quarter’s number and discards the learning that would have improved the next ten. The behaviors that kill culture are the same behaviors that erode organizational memory.
Culture is the context you can’t download. It is built, behavior by behavior, or it is lost the same way.
The lesson for the agentic future is not that we should automate culture; we cannot, and should not try. It is that the human layer comes first, always. AI is the execution layer; it works inside the framework your people set. If that framework is healthy, intelligence compounds. If it is not, you have simply bought a faster way to scale your dysfunction. Any honest account of Enterprise Intelligence has to put culture at the center, not in an appendix.
Predictions are how a manifesto earns the right to be argued with, so here are four I am willing to be held to.
1. The leading enterprise will be self-explaining.
Today, when an executive asks why a target is being missed, they get a dashboard and a scramble: a dozen people assembling, from memory and spreadsheets, a story that should already exist. By 2035 the leading organizations will answer that question the way you answer a question about your own past: instantly, with the reasoning attached. Ask “why are we missing our growth target?” and the organization itself will tell you what changed, why, which assumptions broke, which risks are emerging, what was tried before and how it went, and what it recommends now, with the evidence under each claim. Not because it has more data, but because it finally has structured, intentional context and an intelligence layer that can reason over it.
2. Most of today’s AI initiatives will be quietly written off.
The current wave of enterprise AI spending is, in large part, going to disappoint, and the post-mortems will rhyme. The projects that fail will be the ones that bolted intelligence onto organizations that never built the context to ground it. The projects that succeed will have treated context as the product and AI as the consumer of it, and they will be judged not by their activity but by their outcomes.
3. Enterprise Intelligence will become a category.
Industries form around problems that become too expensive to ignore. Business intelligence became a category when seeing your data became a survival requirement. Enterprise Intelligence, the layer that captures organizational context, connects strategy, performance, and risk, and turns it into compounding, decision-grade intelligence, will become a category for the same reason, on a faster timeline, because AI is turning the Context Crisis from a chronic ache into an acute, quantifiable cost. Categories are not declared by vendors; they are coined by buyers the moment a problem grows too expensive to keep off the budget. That is what is about to happen here. The system of record defined the last era of enterprise software. The system of context will define the next.
4. Context custody will become a named leadership discipline.
The Chief Knowledge Officer faded because it was too early and treated as overhead. The function returns this decade in a more durable form: not a lone librarian, but a federated discipline with real teeth: context ownership embedded in every operating role, a defined cadence for capturing the why, and a senior leader accountable for the standard the way a CFO is accountable for the close. Some organizations will give it a title; the ones that win will give it a culture.
The throughline of all four, in one sentence a skeptic can carry out of the room:
The enterprises that win the next decade will be the ones that learn faster than they forget.
A manifesto that cannot survive contact with a real operating calendar is just decoration. Here is what the argument implies for anyone who runs something, whether or not they ever touch a product like ours.
Treat context as an asset, not exhaust. The reasoning behind your decisions is one of the most valuable things your organization produces, and you are almost certainly discarding it. Capture the why (behind a target, a tradeoff, a killed project, a near-miss) deliberately, while it is fresh, in a form a successor could use. Adopt the Complete Context Standard for your most important metrics and you will be ahead of nearly everyone.
Audit your own forgetting. Ask a concrete question: if your three most knowledgeable people left next month, what would the organization no longer understand about itself? The size of that answer is the size of your exposure, and most leaders have never asked it.
Stop buying information you will not act on. Before the next data source or feed, ask whether the constraint is really that you cannot see, or that you cannot make sense of what you already see. If it is the latter, more information makes it worse. Spend the budget on understanding.
Make context ownership real. Assign it at the metric level, give it a cadence, and hold it to a standard, then have leadership treat completeness the way it treats the financial close. Write for machines as well as people: unambiguous, defined, complete, because your agents will act on exactly what you wrote, at speed.
Protect the culture that makes context possible. Name your Culture Killers and go after them, because every one of them is quietly destroying the organizational memory the next era runs on. Clarity, trust, candor, and a long enough time horizon are not soft virtues; they are the preconditions for an organization that can learn.
None of this requires waiting for a vendor, a budget cycle, or a perfect AI strategy. It requires deciding that understanding your own organization is a first-class job, not a byproduct of staring at charts. The tooling will follow the conviction.
The Industrial Age built management systems to coordinate work. The Information Age built information systems to move data. The agentic age will be built by the organizations that build intelligence systems: systems that capture context, connect it, and turn it into decisions that get better every cycle.
The dashboard was a magnificent answer to the scarcity of information. That scarcity is over. The new scarcity is understanding, and the organizations still pouring resources into seeing more while understanding less are optimizing for a world that no longer exists. Information-rich and intelligence-poor is not a stable place to stand while your competitors learn to climb.
The future does not belong to whoever has the most data, the biggest model, or the prettiest dashboard. It belongs to whoever can turn what their organization knows into what their organization does, reliably, repeatably, and faster than they forget. And as machines take over more of the doing, that turns on one unglamorous discipline: writing the why down clearly enough that an agent cannot mistake it.
Context is the new moat. The future belongs to Enterprise Intelligence.
This paper is an argument, not an advertisement, and the argument stands on its own. But it would be dishonest not to say that I am trying to build toward the future it describes, because that is where the conviction comes from.
RWRD is one attempt, not the only possible one, to build a system of context for the enterprise. It is an intelligence platform that connects what most organizations keep apart: objectives (OKRs), financial and performance metrics, and risk. Put another way, it fuses a system of record (what happened) and a system of recommendation (what to do about it) into a system of context: neither is trustworthy on its own, and what binds them is the connective tissue between them: the why, and the connections that link one number to another. That binding is context, and it is what turns a record you can query and a recommendation you can read into an organization that understands itself. Leadership defines what matters; the human layer always comes first. The platform measures performance continuously, as updates come in, discovers the hidden connections between those domains using network science, surfaces recommendations grounded in your actual data, and lets you simulate decisions before you commit. Define, measure, discover, act, repeat: a flywheel that captures the why and carries it forward. It is built around the Complete Context Standard: today the metrics that matter travel with their descriptions, owners, targets and thresholds, status updates, and action plans attached, and the platform is being extended to close the full learning loop, capturing why a target was missed and whether the fix worked.
Crucially, RWRD is not another ingestion engine. It does not vacuum up the exhaust of your systems of record and hope a model makes sense of it; that is the very approach this paper argues against. It captures the structured layer those systems were never built to hold: the why behind each number, the targets and thresholds it is judged against, the people accountable for it, and the connections between them. And because that context is your moat, you should own it: RWRD is built to be deployed in your own environment, single-tenant or white-label, so your data and your context stay fully under your control, while the integration points to your existing systems get built to your needs, by your team or with ours. The job is never to ingest more data; it is to capture more context, in an environment you trust. It is built for the transition every enterprise now faces: from measuring AI by activity to measuring it by outcomes: decisions improved, risks reduced, revenue created, costs avoided. That is the difference between an agent that acts on noise and one that acts on understanding.
We built it on principles that mirror this paper: the human layer comes first and AI is the execution layer, not the decision-maker; honest accounting over flattering dashboards; and security as a foundation, not a checkbox: SOC 2 Type 2 and GDPR compliant. It is live and in production: free to start as an on-ramp, with single-tenant and white-label deployment plus hands-on deployment and advisory services for the enterprises that need full control of their data.
One proof point is already built into the product itself. The federated model this paper argues for (context owned at the metric level, on a defined cadence, rather than centralized in a single custodian) is how RWRD works in practice: every objective, metric, and risk carries a named owner and an update cadence with automated reminders, so the people closest to each number are the ones who keep its context current.
Whether or not RWRD becomes the system of context your organization uses, I would consider this paper a success if it changes the question you ask: from how to see more to how to understand more, how to stop forgetting, how to make the why as durable as the what. RWRD is our attempt to build that enterprise intelligence layer. We will not get every detail right. But we are convinced the enterprise that wins the next decade will be the one that understands its own context, not merely records its own facts, and we would rather help build that future than watch it arrive. The Context Crisis is real, and the organizations that solve it first will define what comes next. Learn more at rwrd.ai.
- [1] IDC, Worldwide Global DataSphere Forecast; Seagate / IDC, “Data Age 2025: The Digitization of the World From Edge to Core.” Global data creation rising from ~6.5 ZB (2012) to ~181 ZB (2025). seagate.com/our-story/data-age-2025.
- [2] Seagate Technology / IDC, “Rethink Data: Put More of Your Data to Work, From Edge to Cloud” (2020). Survey of 1,500 enterprise leaders; 68% of available business data goes unleveraged. seagate.com (news archive, July 2020).
- [3] McKinsey Global Institute, “The Social Economy: Unlocking Value and Productivity Through Social Technologies” (2012). Knowledge workers spend ~1.8 hours/day (≈9.3 hours/week) searching for and gathering information.
- [4] Gartner, research on data quality (2020–2021): poor data quality costs organizations an average of $12.9 million per year. gartner.com/en/data-analytics/topics/data-quality.
- [5] On NASA’s loss of Apollo / Saturn V institutional knowledge: dispersal of the workforce, dismantling of tooling, and loss of tacit engineering know-how that surviving blueprints alone did not preserve. See NASA APPEL knowledge-management accounts and Lee Hutchinson, “How NASA brought the monster F-1 ‘moon rocket’ engine back to life,” Ars Technica (April 2013), on the reverse-engineering effort that found lost tooling and tacit know-how, not the drawings, were the binding constraint. arstechnica.com.
- [6] MIT NANDA initiative, “The GenAI Divide: State of AI in Business 2025.” ~95% of enterprise generative-AI pilots delivered no measurable P&L impact; the gap was a learning/context gap, not model quality.
- [7] Stanford HAI, “The 2025 AI Index Report.” Inference cost for GPT-3.5-level performance fell ~280-fold (≈ $20.00 → $0.07 per million tokens, Nov 2022–Oct 2024); hardware costs down ~30%/yr and energy efficiency up ~40%/yr. hai.stanford.edu/ai-index/2025-ai-index-report.
- [8] Dario Amodei (CEO, Anthropic), May 2025 remarks: AI could eliminate up to ~50% of entry-level white-collar jobs and drive unemployment to 10–20% within one to five years, concentrated in finance, law, consulting, and tech. Reported by CNN Business and Axios (“white-collar bloodbath”), May 2025.
- [9] Goldman Sachs, “Generative AI Could Raise Global GDP by 7%” and related labor-market research (2023): roughly 300 million jobs globally exposed to automation; about a quarter of work hours in the US and Europe automatable. goldmansachs.com/insights.
- [10] Goldman Sachs Research (Jim Covello), “Gen AI: Too Much Spend, Too Little Benefit?” (June 2024): skepticism that ~$1 trillion of AI capital spending will earn adequate returns, cited here as the cost-skepticism narrative that the falling-cost data contradicts.
- [11] Jeffrey K. Liker, “The Toyota Way” (2004); Toyota Production System: A3 problem-solving reports and the “Five Whys” root-cause method as institutionalized context capture.
- [12] Amazon Web Services, “Correction of Errors (COE)” process: standardized written post-incident analysis (summary, impact, timeline, root cause, corrective actions). aws.amazon.com/blogs/mt (Correction of Errors).
- [13] Betsy Beyer et al. (eds.), “Site Reliability Engineering: How Google Runs Production Systems” (O’Reilly, 2016), “Postmortem Culture” chapter. sre.google/sre-book/postmortem-culture.
- [14] U.S. Army, After-Action Review (AAR) doctrine; U.S. Army Center for Army Lessons Learned (CALL): institutionalized capture, analysis, and dissemination of lessons learned.
- [15] Thomas H. Davenport and Laurence Prusak, “Working Knowledge: How Organizations Manage What They Know” (Harvard Business School Press, 1998), on the 1990s knowledge-management movement; and CIO and Harvard Business Review retrospectives on the decline of the Chief Knowledge Officer role amid unclear ROI and over-centralization.
- [16] Keele Advisory (Tracey Keele), “Culture Killers” series and the working definition of culture as “how work gets done” and “the worst behaviors we tolerate.” keeleadvisory.com.
- [17] McKinsey & Company, research on transformation failure rates (~70%) and “The State of Organizations 2026” (≈75% of organizations struggle to build high-performance cultures); culture cited as the most common cause of failure.
- [18] U.S. House Committee on Transportation & Infrastructure, “Final Committee Report: The Design, Development & Certification of the Boeing 737 MAX” (Sept. 2020); subsequent expert-panel reviews of Boeing’s safety culture (“culture of concealment”; erosion under cost and schedule pressure).
Chris Harding is the founder of RWRD, an enterprise intelligence platform that merges strategy, performance, and risk into one interconnected narrative for the agentic future. He has spent over two decades at the intersection of strategy, risk, and execution: in strategy and risk consulting at KPMG across South Africa, the US, and the UK; building the enterprise risk function at LendingClub on its path to becoming an OCC-regulated bank; and leading strategic AI adoption, governance, enterprise risk, and technology compliance at AppFolio. He is a Certified Internal Auditor and holds a BCom (Honours) in Accounting and Internal Audit from the University of Pretoria. He believes every metric has a story, and that the organizations that learn to tell those stories together will win the next decade.
Connect: linkedin.com/in/chris-harding-1b30541b · rwrd.ai
