In January, CLA reported its AI agent

now does the work of 853 full-time

employees and has saved the company $60

million. In the same earnings cycle, its

CEO admitted publicly that the AI

strategy had cost something far more

valuable than $60 million, and he's

still trying to buy it back. This is not

another AI is overhyped story. It is

actually the opposite. The AI work too

well. And the distinction between AI

that fails and AI that succeeds at the

wrong thing is the most important

unsolved problem in enterprise AI right

now. This is a bigger problem than

context engineering per se, although

that's a piece of it. It's bigger than

prompt engineering, which now frankly

looks like a warm-up act. I'm going to

call what we're talking about here

intent engineering. It is the discipline

of making organizational purpose like

goals, values, tradeoffs, decision

boundaries. These need to become machine

readable and machine actionable so that

when you deploy an autonomous system, it

optimizes for what your company actually

needs, not just what it can measure.

Here's the CLA backstory. In early 2024,

Ara rolled out an AI powered customer

service agent. It handled 2.3 million

conversations in the first month across

23 markets in 35 languages. Resolution

times dropped from 11 minutes to two.

The CEO projected $40 million in

savings. And then customers started

complaining. Generic answers, robotic

tone, no ability to handle anything

requiring judgment. By mid 2025, CEO

Sebastian Seycowski told Bloomberg that

while cost was a predominant evaluation

factor, the result was lower quality.

End quote. Clarin began frantically

rehiring the human agents it had gutted.

Most people tell this story as proof

that AI can't handle nuance, and that

was a comforting reading in early 2025.

A more interesting reading in 2026 is

that the AI agent was extraordinarily

good at resolving tickets fast and that

that was the wrong goal to give the

agent. Clara's organizational intent

wasn't resolve tickets fast. It was

actually build lasting customer

relationships that drive lifetime value

in a very competitive fintech market.

Those are profoundly different goals and

they require profoundly different

decision-making at the point of

interaction. A human agent with five

years at the company knows this

difference intuitively. She knows when

to bend a policy, when to spend three

extra minutes because the customer's

tone says they're about to churn, when

efficiency is the right move versus when

generosity is the right move. She knows

this because she absorbed Clara's real

values. Not the ones on the website, but

the ones encoded in the decisions

managers make every day in the stories

veterans tell new hires. In the

unwritten rules about which metrics

leadership actually cares about when

Bush comes to shove, the AI agent knew

none of it. It had a prompt. It had

context. It did not have intent. I am

concerned that the AI agent

inadvertently

reflected the real values of CLA when it

behaved the way it did because the real

values of CLA may have been to save the

money first. Nevertheless, customers

pushed back and called CLA back toward

its own stated values. And that's been a

good thing because the $60 million in

savings since the program rolled out

have been not nearly enough to cover the

reputational damage to CLA from becoming

a public laughingtock over AI and from

aggressively pushing customer service

resolutions that did not meet customer

needs. Ironically, rolling out an AI

without much thought, but perhaps in

line with organizational directives, did

more to push CLA toward its own stated

values than perhaps anything else would

have done. I want to be precise about

this story. The point here is not to

talk about CLA per se. The point is to

talk about what our organizational

intent is with AI agents, what our goals

are, and how agents need to reflect the

larger perspective and longer term

challenge of our organizations as they

become more sophisticated and operate

for longer time scales themselves. We

have agents now that run for multiple

weeks. We have agents soon that will run

for multiple months. We are at a level

where it is time to think about how

agents interact with organizational

goals very seriously and CLA is just

kind of an example of why that's

important. I want to be precise about

how we got here because naming things

matters. Naming is how we create a

shared understanding and I think we are

short on naming things correctly when it

comes to intent and context. Prompt

engineering was the first discipline in

the age of AI. It was individual,

synchronous, and sessionbased. You sit

in front of the chat window, you craft

an instruction, you iterate the output.

It's a personal skill, and the value is

personal. This is the era that produced

a thousand how to write the perfect

prompt blog post. Most of them are

terrible. Context engineering followed

prompt engineering. It's the one the

industry is currently grappling with.

Anthropic published a foundational piece

in September of 2025 that defined

context engineering is the shift from

crafting isolated instructions to

crafting the entire information state

that an AI system operates within.

Chains Harrison Chase put it more

bluntly in a Sequoia Capital interview

when he described it as everything's

context engineering. Context engineering

is such a good term. I wish I came up

with that term because it describes

everything we've done at Langchain

without knowing the term existed. End

quote. That's pretty good. Context

engineering is where the action is right

now. Building rag pipelines, wiring up

MCP servers, structuring organizational

knowledge so agents can access it. It's

necessary, but it's not sufficient. And

I think the industry is about to

discover that in a very expensive way.

Intent engineering. Intent engineering

is the third discipline and it's the one

that almost nobody's building for yet.

Context engineering tells the agents

what to know. Intent engineering tells

agents what to want. It's the practice

of encoding organizational purpose into

infrastructure, not as pros in a system

prompt, but as structured, actionable

parameters that shape how agents make

decisions autonomously. It's the layer

that would have told Clara's AI agent,

"Yes, you can resolve this ticket in 90

seconds, but the customer has been with

us for years, and their tone indicates

frustration. Spend the extra time. Offer

them a specialist." The goal is

retention. Without intent engineering,

you get what Claragot, a technically

brilliant agent optimizing for exactly

the wrong objective. You get what

Deoid's 2026 state of AI in the

enterprise report found across 3,000

some leaders in 24 countries. 84% of

companies have not redesigned jobs

around AI capabilities and only 21% have

a mature model for agent governance.

These numbers aren't a technology story.

They're an intent failure. The models

work. The context pipelines are getting

better. What's missing is the

organizational infrastructure that

connects AI capability to organizational

purpose. I cited the failure stats from

deote above. I want to show you the

other side of the ledger because the

juxtaposition makes this all very

disorienting. Investment in AI continues

to be massive and accelerating. Deoid's

tech value survey found that 57% of

respondents were putting between 21 and

50% of their digital transformation

budgets into AI automation. and 20% of

companies were investing over half on

average $700 million for a company with

13 billion in revenue. KPMG's Q4 AI

pulse survey showed capital flowing ROI

confidence rising in agents moving from

pilots to professionalized platforms.

Gartner is predicting that by 2028 15%

of dayto-day work decisions will be made

autonomously by agents. I think that

might be low. So the money is real, the

deployments are real, and the results

are very much in between. 74% of

companies globally report they have yet

to see tangible value from AI. McKenzie

found 30% of AI pilots failed to achieve

scaled impact. These numbers all coexist

together with the investment numbers.

There's not really a contradiction here

if you start to peel the onion and

understand things more carefully. What

we're describing when we talk about a

pattern of scaled investment and

somewhat mixed results on deployment is

that organizations have solved can AI do

this task at an individual task level

and they have completely failed to solve

can AI do this task in a way that serves

our organizational goals at scale with

appropriate judgment. That second

question that's an intent engineering

question. Look at what happened with

Microsoft Copilot. One of the most

heavily invested enterprise AI products

in history. Microsoft poured billions

into infrastructure, embedded AI into

every office application, and launched

an aggressive enterprise sales campaign.

85% of Fortune 500 companies adopted it,

and the adoption stalled hard. Gartner

found that only 5% of organizations

moved from a co-pilot pilot to a larger

scale deployment. Only about 3% of the

total Microsoft 365 user base actually

adopted Copilot as paid users. Bloomberg

reported Microsoft slashing internal

sales targets after the majority of

salespeople missed their goals. Even

inside companies that signed six figure

co-pilot deals, employees resisted.

Reddit threads are full of engineers at

multi-billion dollar companies

describing their organizations

downgrading licenses because employees

preferred another AI. maybe chat GPT,

maybe Claude. The standard explanation

for co-pilot struggles centers on UX

problems and model quality. And those

are definitely real issues, but they're

not the fundamental issue. The

fundamental issue is that deploying an

AI tool across an organization without

organizational intent alignment is like

hiring 40,000 new employees and never

telling them what the company does, what

it values, or how to make decisions. You

get lots of activity and not much

productivity. You get AI usage metrics

in a dashboard and almost no measurable

impact on what the organization is

trying to accomplish. That's not a tools

problem. That's an intent gap. I want to

get structural because vague handwaving

about AI transformation is exactly what

we're trying to avoid here and what so

many organizations get into trouble

doing. There is an intent gap today and

it operates across three distinct

layers, each one at a different

altitude. Getting any one of them right

is helpful. Getting all three right is

the difference between having AI tools

and having an AI native organization.

Layer one is what I'm going to call a

unified context infrastructure. This is

the layer the industry is most aware of

and it's still not really built yet.

Right now, every team building agents

rolls their own context stack. One team

pipes Slack data through a custom rag

pipeline. Another manually exports

Google Docs into a vector store. A third

built an MCP server that connects to

Salesforce but not to Jira. A fourth

team doesn't know the other three exist

yet. This is what one analyst called the

shadow agents problem and it mirrors the

shadow IT crisis of the early cloud era

except the stakes are much higher

because agents don't just access data,

they act upon it. Security and

compliance teams can't allow arbitrary

unvetted agents running on developer

laptops to access critical systems like

customer PII, financial data or

healthcare records. But without

sanctioned infrastructure, that is

exactly what is happening. The model

context protocol which Anthropic

introduced late in 2024 and donated to

the Linux Foundation in December of 2025

is the most promising attempt at

standardization at this point. UCP has

seen a ton of adoption. OpenAI, Google,

Microsoft, and more than 50 enterprise

partners have committed to it. It's

become the de facto standard. Monthly

SDK downloads are close to 100 million

now, I think. But protocol adoption and

organizational implementation are very

different things. Having a USBC standard

does not help if your company hasn't

decided which ports to install, who

maintains them, or what gets plugged in.

The context infrastructure question is

not really a technical question. You can

configure MCP servers. It is

architectural and political. Which

systems become agent accessible? Who

decides what context an agent can see

across departments? How do you version

organizational knowledge so agents

aren't operating on stale information?

How do you handle the fact that the

sales team Slack context and the

engineering team Slack context encode

completely different institutional

assumptions? Deoid's 2025 survey found

that nearly half of organizations cited

data searchability and data reusability

as top challenges blocking AI

automation. I'm surprised that number

isn't higher. As their analyst put it,

the shift required is from traditional

ETL data pipelines to enterprise search

and indexing, similar to how Google made

the worldwide web discover. The data

does exist inside corporations. The

agents also exist increasingly, but the

connective tissue between them, the

organizational context layer and the

structures and safeguards to ensure

that's accessed correctly, that mostly

doesn't exist. Now, we're going to move

on to layer two, the coherent AI worker

toolkit. So, everyone's rolling out

their own AI workflow. One person uses

Claude for research and chat GPT for

drafting. Another uses cursor for code

and perplexity for factchecking. A third

has built a custom agent chain using

langraph. A fourth is copy pasting into

a chat window. None of these employees

can articulate their workflow in a way

that's transferable, measurable, or

improvable by anybody else. And this

matters because the difference between

individual AI use and organizational AI

leverage is enormous. It's the

difference between having one good hire

and having a system that makes everybody

better. It's the difference I've been

writing about for a year between AI

activity and AI fluency. The former has

30% gains that you get from bolting AI

onto existing workflows. And the latter

has the 300% gains you get from

rethinking the workflow itself around AI

capabilities. But here's what we need to

realize. Fluency doesn't scale through

training alone. It scales through shared

infrastructure. Whether any individual

person has Slack doesn't matter. Whether

an agent can search 50 people's Slack

context plus their docs plus their

project plan plus the customer data,

that's what determines whether the agent

can do organizational scale work rather

than individual scale tasks. Lloyd's

2026 report found that workforce access

to sanctioned AI tools expanded by 50%

in a year. But that doesn't mean that

access is sufficient. Organizations are

often giving people tools without giving

them or their agents the organizational

context and data that allow those tools

to deliver real value. And that's where

Clara's story intersects with co-pilot

story. Tools deployed without

organizational infrastructure become

very expensive toys. The 74% of

companies reporting no tangible value

from AI are probably not failing because

of models. They're failing because

there's no shared understanding of how

AI tools connect to organizational

workflows, of where AI automation should

replace human effort, of where it should

augment it, of where human judgment

should be non-negotiable, all the things

that CLA should have done. All the

things that the co-pilot salespeople

didn't tell you about. This is the issue

today with AI in the enterprise. We are

not taking that data and context layer

seriously. And that doesn't allow us to

even approach layer three, which we're

going to talk about next. Intent

engineering proper. This is the layer

that almost certainly doesn't exist in

your business. It's the one I think

matters the most, and it requires

something genuinely new. OKRs were

designed for people. They encode human

readable goals. They assume human

judgment about prioritization,

trade-offs, values, and exceptions. They

assume a manager can look a direct

report in the eye and say, "Here's what

matters this quarter." and trust that

the report will interpret that guidance

through a mesh of institutional context,

professional norms, and personal

judgment developed over months and

years. Agents don't have any of that. An

agent does not know your company's OKRs

unless you put them in the context

window. It doesn't know which trade-offs

your leadership team would prefer unless

you encode those preferences in a way it

can act on. It doesn't know the

difference between a decision that

should be escalated and one it should

make autonomously unless you define the

boundary. And unlike a human employee,

an agent will not absorb your company

culture through osmosis for 6 months,

through all hands meetings, through

hallway conversations, and through

watching senior people handle ambiguous

situations. When a human employee joins

a company, alignment happens through a

100 informal mechanisms. You read the

wiki, you have a slack chat, you develop

judgment, you have a happy hour with

someone. None of that works for agents.

Agents need explicit alignment, and they

need it before they start working, not 6

months after. This means organizations

need to develop something that mostly

doesn't exist. Machine readable

expressions of organizational intent.

Think about what that requires. It is

not just put the OKRs in the prompt. It

is a cascade of specificity that most

organizations have never had to produce

because humans could fill in the gap. It

is a cascade of specificity that

organizations have never had to produce

because humans could fill in the gaps.

At the top, you need goal structures

that agents can interpret and act on,

not increase customer satisfaction.

That's a human readable aspiration. You

need an agent actionable objective. An

agent needs to know what signals

indicate customer satisfaction in our

context. What data sources contain those

signals? What actions am I authorized to

take to improve them? What trade-offs am

I empowered to make? speed versus

thoroughess, cost versus quality, and

where are the hard boundaries I may not

cross. Below that, you need what I would

call delegation frameworks, tenants

translated into decision boundaries.

Amazon's leadership principles work for

humans because humans can interpret

customer obsession through contextual

judgment. An agent needs that principle

to be decomposed. When customer request

X conflicts with policy Y, here is the

resolution hierarchy. When data suggests

action A, but the customer expressed

preference B, here's the decision logic.

These are not rules in the traditional

sense. They're encoded judgment. The

kind of organizational knowledge that a

senior employee carries in her head

after 5 years and a new hire will absorb

gradually. Agents need it now. And at

the base, you need feedback mechanisms

that actually close the loop. When an

agent makes a decision, was it aligned

with organizational intent? How do we

know? This is exactly what happened at

CLA. The agent optimized for resolution

speed because that was the objective it

could measure. Nobody had encoded the

objectives that mattered most.

Relationship quality, brand trust,

customer lifetime value, the contextual

judgment about when to be efficient and

when to be generous. Those objectives

lived in the heads of the human agents

who had to walk out the door because

they were fired. The age of humans just

know is ending. Intent engineering is

the discipline of making what humans

know explicit, structured, and machine

actionable. Not because the humans are

leaving, although some of them will, but

because the agents arriving to work

alongside the people cannot function

without it. If there is anything I want

you to take away from this video, it is

not if I can do this intent engineering,

I can get rid of the people. You should

be regarding agents as rather

undependable actors and recognizing that

you need humans to both encode intent

engineering and maintain successful

agentic systems that scale. That's how

you actually start to drive agents in

production. So why hasn't this been

built yet? First, it's genuinely new.

Before agents could run autonomously

over long time horizons, we did not need

this. The human was the intent layer.

The agent never needed to understand

organizational intent because you were

standing right there. Longrunning agents

break the model and demand a new way of

thinking. And that's what this video is

about. Second, the people who understand

organizational strategy like executives

are not the people who build agents. And

the people building agents like

engineers are not the people who

understand organizational strategy very

frequently. This is a classic two

cultures problem. And it's acute in AI

because the technology is moving so fast

that the organizational thinkers cannot

keep up and the technologists, they

don't think it's their job. MIT found

that AI investment is still viewed

primarily as a tech challenge for the

CIO rather than a business issue that

requires leadership across the

organization. That framing that's going

to guarantee an intent gap that has real

implications for your AI agents. CIOS

can build infrastructure, but intent

comes from the entire leadership team

working together. The people who

actually decide what the organization

values and how it makes trade-offs need

to be talking with engineering more.

Third, just really hard. Making

organizational intent explicit and

structured is extremely difficult. Most

organizations have never had to do this.

Their goals live in slide decks, in OKR

documents that get half read and

referenced at personal reviews once a

year, in leadership principles that get

cited in performance reviews, but really

they don't get operationalized in the

tacet knowledge of experienced employees

who know what to do in ambiguous

situations even though they've never

been told. Nobody has strong muscles

here because most organizations have

never exercised them. So what does a

solution look like? I don't want to just

leave you with a gap. First, at the

infrastructure level, you need to

develop a composable vendor agnostic

architecture that enables agents to

operate across systems, tools, and

models securely and at scale. MCP is a

sample protocol layer for this. But the

organizational implementation requires

decisions about data governance, access

controls, freshness guarantees, and

semantic consistency that no one

protocol is going to make for you. The

companies that build this well will

treat it like they treated their data

warehouse strategy in the as a core

strategic investment not just an IT

project. At the workflow level you need

what I would call an organizational

capability map for AI. A shared living

understanding of which workflows are

agent ready which are agent augmented

with human in the loop and which remain

human only. This is not a static

document that gets filed in confluence

and dumped. It's an operating system

that evolves as agent capabilities keep

improving and as organizational context

infrastructure matures. The companies

that do this well are likely going to be

creating a new role. It will be called

something like an AI workflow architect

and it will sit between engineering

operations and strategy and that person

is going to be very busy at the

alignment level. You need the genuinely

new thing. Goal translation

infrastructure that converts human

readable organizational objectives into

agent actionable parameters. This

includes decision boundaries,

escalation, value hierarchies like how

the agent resolves trade-offs and

feedback loops. How you measure and

correct alignment drift over time.

Google's agent development kit is one of

the earliest attempts to formalize this

at a technical level. It separates agent

context into distinct layers. working

context, session memory, long-term

memory, and artifacts. Each of these has

specific governance. There's also

emerging academic work. A recent paper

from researchers at Google DeepMind

proposed five levels of AI agent

autonomy. Operator, collaborator,

consultant, approver, and observer, each

with different intent alignment

requirements and different human

oversight models. These are just the

early sketches. The integrated system is

really whites space. Building context

infrastructure plus workflow mapping

plus intent alignment is new and it's an

enormous challenge. If OKRs were the

management innovation that let Intel

align thousands of humans to shared

objectives in the 70s, intent

engineering is the management innovation

that lets organizations align hundreds

or thousands or tens of thousands of

agents to those same objectives in 2026.

While those agents operate at speeds and

scales, no human manager can supervise.

The parallel is direct. The urgency has

never been greater. OKRs have taken

decades to become standard management

practice. We do not have 20 years to

wait. For the past 3 years, the AI race

has been framed as an intelligence race.

Who has the best model? Who tops the

best benchmarks? Who has the biggest

context window? That framing made sense

when models were a bottleneck. But

models are not the bottleneck today. Not

for most organizational use cases. The

frontier models like Opus 4.6 or Gemini

3 or GPT 5.2. These are all

extraordinarily capable models. The

differences between them matter far less

than the differences between

organizations that give them clear,

structured, goal-igned intent and

organizations that don't. The race is an

intent race. Not who has the smartest AI

in their systems, but who has built the

organizational infrastructure that lets

AI operate with the fullest, most

accurate, most strategically correct

understanding of what the organization

is trying to accomplish. The company

with a mediocre model and extraordinary

organizational intent infrastructure

will outperform the company with a

frontier model and fragmented,

inaccessible, unaligned organizational

knowledge every single time. This means

that the most important AI investment in

2026 isn't really a model subscription.

It's not another co-pilot license. It's

organizational intent architecture.

Making your company's goals, values,

decision frameworks, and trade-off

hierarchies discoverable, structured,

and agent actionable. It's building the

alignment infrastructure that lets

agents make decisions that aren't just

technically correct, but that are

strategically coherent. It's developing

the shared language and shared systems

that let AI capabilities scale from one

heroic engineer to 40,000 knowledge

workers operating in concert. Clara's

story was not AI doesn't work. The AI

worked brilliantly. That was the

problem. It was so good at optimizing

for the measurable objective that nobody

noticed it was destroying the ones that

really mattered. Trust. The 700 human

agents that got laid off took with them

the institutional knowledge that really

mattered. The knowledge that had never

been documented. Humans just knew. The

lesson is to build the intent layer so

that agents don't need to absorb

organizational values through osmosis

because they can't. The lesson is to

recognize that agents need humans

working alongside them. Maybe Clara has

finally learned that the prompt

engineering era asked, "How do I talk to

AI?" The context engineering era is

asking us now, "What does AI need to

know?" And the intent engineering era is

beginning to ask us the question that

really matters. What does the

organization need AI to want to be

productive? Context without intent is

like a loaded weapon with no target.

We've spent years building AI systems.

2026 is a year when we learn to aim them

toward an organizational intent that

really matters. If you're listening to

this, you are involved at one or more of

these layers. Everybody is from the

individual contributors who are working

against these systems and practicing

prompting and trying to use them to

gather context all the way up to the

systems designers and all the way up to

the seauite. It is up to all of us to

build layers that enable agents to act

productively in line with organizational

values. If we are not careful, failure

to do so is going to lead to AI agents

that cause active harm to the business.

That's what Clarno learned. Don't do

that. Build systems that encode both

context and intent at the organizational

scale. The clock is running. And the

teams that do this are going to be able

to unleash the power of the agents that

are running for weeks and soon for

months with a whole lot more confidence

than the people who are building systems

where they don't encode intent, where

they don't encode values, where they

don't encode tradeoffs, where you cannot

trust an agent not to hang up on a

customer just because you told them to

make the call shorter. Don't do that.

Build for long-term intent because

agents with long-term intent are