H1 Trends Brief · H1 2026 Powered by Watchtower
When AI becomes an operating model.
The demand now is better, faster, and cheaper, no longer one at the expense of the others. Andrew Au · Co-CEO, Intercept
- Workflow design is now strategy. Token costs are climbing as the subsidies fade, and the answer is right-sizing which model handles what. One company spent $500M on AI in a single month. Uber blew its 2026 budget by April.
- "Agent" is being used before anyone agrees what it is. Only 14% of marketers have configured prebuilt agents. The rest are calling everything an agent, assuming IT owns it, or admitting they do not really know.
- 59% of buying committees will demand more proof in 2026. Half say the biggest change ahead is the need for content AI can actually read. The work is two-track: reformat for today's models, and feed the next training run.
At a glance · The six-pillar radar
Where the six pillars stand in H1 2026.
Jump to any pillar to read the full section, or scan below for the 30-second version. Job Confidence is the outlier and the most revealing number in the set.
Individual AI Proficiency
Self-taught, one person at a time. Value stays locked inside individuals.
↑ rebound after three declinesOrganizational AI Readiness
Nobody owns it yet. Progress stalls wherever ownership is unclear.
↑ modest riseJob Confidence
The exception in the rebound. Discomfort with what comes next.
↓ lowest reading in the seriesSentiment Toward AI
Adopted faster than it is trusted. Consistency is how trust gets earned.
↑ inside marketingOrganizational AI Adoption
Workflow design becomes strategy. Token cost is the new hard constraint.
↑ uneven but realFuture Outlook
Recommendation is the new ranking. Two fronts to experiment on.
↑ highest in the setChapter 1
Where things stand.
The AI conversation has changed. Leadership stopped asking about adoption and started asking about return. Before diving into the data, three sections of context: the shift, the executive summary, and what history says about what happens next.
Sections 01, 02, 03The question is no longer whether to adopt AI.
The conversation about AI continues to change. The question in the room is no longer whether to adopt it, but whether it is working, driving impact, and becoming part of a system the team can reliably run. Leadership wants to see the return.
We have been here before. In 1990, with companies pouring money into computers and seeing little of it reflected in productivity numbers, Michael Hammer made the case in the Harvard Business Review that the technology was not the problem. Companies were using computers to speed up the way they already worked, paving the cow paths instead of building new roads. The returns only arrived when they redesigned the work around the technology and built a more disciplined way to run it. Ford rebuilt its accounts payable process and cut headcount by roughly three quarters, where automating the old process would have saved a fifth.
The same logic governs AI now. The teams adding a copilot here and a chatbot there, on top of the workflow they already had, keep waiting for the return. The teams redesigning the work around AI, and turning that redesign into a repeatable operating model, are the ones who will see it.
This is the fourth report in our series on how AI is changing B2B marketing, and it captures the biggest shift yet.
AI has moved from something teams use to something teams increasingly run the function on, which makes workflow design and governance part of the work itself.
A year ago, leadership cared about adoption. About which tools the team had, how many people used them, and how fast it was moving. Today, leadership cares about outcomes, and about whether the system producing them can be trusted.
This report lays out where that shift appears in the data, what it means for how marketing teams are run, and what a more deliberate operating model now requires from leadership.
AI stops sitting on the sidelines.
Adoption is up again this quarter, but the more important change is what marketers now expect from AI, and from the systems forming around it. Acceptable output is cheap to produce, so volume no longer impresses anyone. The bar has shifted to whether AI can match the quality a strong team would produce while costing less and moving faster.
17%
Marketers have stopped measuring AI by whether they are using the tools and started measuring it by whether it improves the business.
The conversation about agents is running well ahead of the systems needed to define, govern, and manage them. Teams use the word "agents" freely, while few have agreed internally on what an agent is, who builds one, or who is responsible for it when it does the wrong thing.
More broadly, AI concerns have moved in the same direction. A year ago, they were ethical and reputational, about copyright, hallucinations, and brand drift. Now they are operational, centered on whether teams have a reliable way to manage what AI is doing inside the funnel.
14%
These shifts share one thing in common. When execution becomes abundant, judgment becomes the scarce resource, and the teams that win are the ones keeping the tightest collaboration between people and AI. That is what it means as AI moves from the edge of the marketing function into its operating model: the advantage shifts to teams that can build a repeatable system for judgment, execution, and accountability.
The work changes when the system changes.
The companies that won the last technology shift were the ones that redesigned the work rather than digitizing the old version of it, then made that redesign durable through the way the work was actually run. The same will hold for AI. Different this time is speed, because the 1990s redesign that took the better part of a decade is now expected inside a planning cycle.
"The gap between what AI can do and what most teams actually get out of it is not a technology problem. It is a design problem, and right now it is the most valuable problem in marketing. The teams that win the next few years will be the ones that stop bolting AI onto the old workflow and start rebuilding the workflow around it."
Shaheen Yazdani · Co-CEO, Intercept
Chapter 2
The six-pillar radar.
The Watchtower radar tracks six pillars quarter to quarter. Five of them rose this cycle. Job Confidence fell. This section is the overview before the six deep dives that follow.
Section 04Six signals that show where we're heading.
Five of the six radar dimensions rose in the first quarter of 2026. The sixth, Job Confidence, fell, and it is the most revealing number in the set.
- Individual AI Proficiency: 3.4 of 5, up from 3.17 last quarter — rebound after three declines
- Organizational AI Readiness: 3.1 of 5, up from 3.02 last quarter — modest rise
- Sentiment Toward AI: 3.4 of 5, up from 2.86 last quarter — inside marketing
- Organizational AI Adoption: 3.3 of 5, up from 3.22 last quarter — uneven but real
- Future Outlook: 3.5 of 5, up from 2.98 last quarter — highest in the set
- Job Confidence: 2.7 of 5, down from 2.85 last quarter — lowest reading in the series, the only pillar to fall
Bring this into the work
Want to see this operating model in practice?
Intercept Labs co-invests with clients on the experiments that turn radar signals into deployed workflows.
Chapter 3
Underneath each signal.
Six pillars, one per section. Each opens with a 30-second summary so you can skim the sections you want to read in full. Pillar 3 is the outlier and worth the extra attention.
Sections 05 through 10In 30 seconds
Proficiency climbed back to 3.4 after three straight declines. The rise is self-taught, use-case-specific, and locked inside individuals. Enterprise value stays out of reach until the skill becomes a shared capability.
Individual AI Proficiency climbed back to 3.4 this quarter after three straight reports of decline. As a general observation, AI is being used as a point solution rather than a shared capability, picked up ad hoc by the people motivated enough to teach themselves, rather than encoded into how the team works. The result is pockets of real productivity that never add up to enterprise value, because what one person figures out rarely reaches the person next to them.
The Rebound · Four-Quarter Trajectory
Individual AI Proficiency over four quarters
(rebound trajectory placeholder)
Almost all of this proficiency is self-taught, since structured AI training inside most organizations is still largely missing. The value that does show up, in social content, workflow automation, SEO content, and AI search optimization, stays locked inside those few people and select use cases.
The Signal
AI lands first where the work is repetitive, fast, and modular, which is also where confidence is highest. Confidence drops sharply on anything that crosses team boundaries or asks the tool to exercise judgment, which is exactly where the harder enterprise value sits.
The Implication
The fact that this skill is self-taught says as much about motivation as it does about ability. People are teaching themselves AI because they have a reason to, but self-teaching alone does not create the shared workflows, standards, or repeatability needed for organizational value.
"AI is a reset moment for how people learn, and it rewards a growth mindset. Mindset is fast becoming the real divide."
Shaheen Yazdani · Co-CEO, Intercept
In 30 seconds
Readiness sits at 3.1. The gap is not tools. It is coordination: prioritizing use cases (12.1%) and strategic clarity (8.5%). Until a named owner runs it, the score stays stuck.
"Organizational AI readiness was never about turning a few people into power users. It is about lifting the whole organization, and that takes executive vision, a culture that builds a growth mindset into the everyday, and the infrastructure underneath it all. This is an organizational transformation, not a tooling upgrade."
David Toto · Managing Director, Intercept
Organizational AI Readiness sits at 3.1 this quarter. Watchtower defines readiness as structural preparedness: the training, governance, and leadership alignment behind running AI. The data says that is the gap. The harder friction is not just getting people to use AI. It is coordinating it, prioritizing the use cases, creating strategic clarity, and putting a workable system around it across the function. The two most-reported blockers this quarter were difficulty prioritizing AI use cases (12.1%) and a lack of strategic clarity (8.5%). Both are management problems rather than technology ones.
Same Goal · Three Different Bets
How organizations are approaching AI ownership
Bet 02 · Central Control — lock everything behind a central AI function
Bet 03 · Data First — fix data estate and infrastructure so the org is ready when AI scales
The common thread across all three is that progress stalls wherever ownership is unclear. Until a named owner decides how AI gets prioritized, governed, and resourced, there is no real operating system for the work, only scattered experimentation, and the readiness score stays stuck.
In 30 seconds
Job Confidence dropped to 2.7, the lowest reading across four reports and the only pillar to fall this quarter. The gap between conviction in the tech and confidence in personal relevance is the clearest signal in the data.
Job Confidence fell to 2.7 this quarter, its lowest point across the four reports, and it is the only dimension that dropped while every other one rose. That gap, between conviction in the technology and confidence in personal relevance, is the clearest signal in the data.
All Six Pillars · Q4 2025 to Q1 2026
Five pillars rose across the quarter. Job Confidence fell.
The Signal
Underneath the number is a broad discomfort with what comes next, and most of it is simply not knowing. The frontier labs have not been specific about how any of this benefits the people doing the work, and the message that comes through is a vague reassurance that it will all work itself out.
The Implication
Executives owe their teams a concrete account of what AI does to the labor market and to specific roles, and a real plan for reskilling, retraining, and how jobs evolve rather than simply disappear. The conversation that actually settles people is about new value creation and where the work goes next, more than about efficiency.
"Using AI to chase efficiency is short-sighted. The real prize is new value creation, doing the things we could never do before. Keep doing the same work faster and all you get is 'ghost GDP.' Point it at new ideas and new possibilities, and the opportunity in front of us is abundant."
Shaheen Yazdani · Co-CEO, Intercept
In 30 seconds
Marketing sentiment is 3.4. Public sentiment is far colder: 26% positive, 46% negative. Trust follows consistency. Inconsistent AI tooling breaks trust before governance does.
Sentiment Toward AI sits at 3.4 inside the marketing teams we track, but the public mood is far colder.
Inside vs Outside
Two very different pictures of AI
Public: 26% positive / 46% negative (NBC News) · 70%+ say AI is moving too fast (Economist / YouGov)
Across North America this spring, commencement speakers who praised AI were booed off the stage, from Eric Schmidt at the University of Arizona to a speaker at the University of Central Florida who called it the next industrial revolution. NPR ran a segment advising 2026 speakers not to raise the subject at all.
The Signal
Inside the teams using AI every day, the same split shows up. People who get reliable results from AI come to trust it, and people fighting tools that work brilliantly one day and break the next do not.
The Implication
Trust in AI grows when the results are consistent. That comes from building AI into a reliable operating system for the work, not from leaving outcomes to individual prompting habits.
Watchtower Signal Meter
Where trust breaks first this quarter
Poor integration reliability · 12%
Procurement slows adoption · 12%
Automation causes errors · 11%
Agents take wrong actions · 11%
"Treat AI as a point solution and the experience swings wildly from one person to the next. Build agentic systems that give people consistent, reliable outputs without the guesswork, and trust starts to follow. Consistency is how you earn it."
Francis Silva · Chief Technology Officer, Intercept
In 30 seconds
Adoption is real (3.3) but uneven. 36% have AI embedded in day-to-day workflows; 13% run agentic workflows in production. The bigger shift this quarter is cost: token rates are climbing as subsidies fade.
Organizational AI Adoption sits at 3.3 this quarter. The question is no longer whether to do this, but which of the existing pilots are solid enough to become part of the workflow infrastructure that actually runs the business. Adoption is real, but uneven.
Adoption Depth · Where Workflows Actually Run
Embedded vs. agentic, this quarter
Agentic workflows running in production · 13%
The bigger shift this quarter is the cost of running it all.
The Cost Reality
Cheap tokens were subsidized to drive adoption. As the subsidy fades, the rates are climbing.
April when Uber burned through its entire 2026 AI budget, with its own COO questioning whether the spend was producing better products
Cost governance was an afterthought a year ago, and is becoming a hard constraint on how far adoption can go. Some teams are already pulling back on AI subscriptions.
The Signal
Teams have moved on to designing the workflows AI runs inside, and at the same time they are waking up to what those workflows cost. The token rates that made heavy experimentation painless were subsidized to win adoption, and they are going up.
The Implication
The next phase of adoption is about efficiency as much as capability. Design workflows that match the model to the job rather than routing everything through a frontier model. Keep routine tasks on smaller or on-device models and save the expensive frontier calls for the work that truly needs them.
"A lot of what we are doing today is the equivalent of taking a Ferrari to drive one block. You do not need a frontier model to rename a file."
Andrew Au · Co-CEO, Intercept
In 30 seconds
Future Outlook at 3.5, the strongest reading. LLMs are the first gate a buyer passes through. 59% expect their committee to demand more proof, and 50% say the biggest change is content AI can actually read.
Future Outlook sits at 3.5 this quarter, the strongest reading in the set, because marketers can see where buying is going even if they cannot yet control it. More of the early buying journey now runs through an LLM before a human is involved, and the question consuming marketers is how to get recommended when it does.
59%
The Signal
Almost everyone is now trying to optimize for how often the models recommend them, because the LLM has become the first gate a buyer passes through. Nobody has a settled playbook for it, and the ground keeps moving, since every new model release can change which brands get surfaced and why.
The Implication
This rewards rapid, disciplined experimentation on two fronts at once.
Two-Front Experimentation
Optimize for today's models. Feed the next training run.
Front 02 · The Next Training Run: be surfaced by default through proof, structure, verifiable claims, earned coverage
"The market is moving toward proof, structure, and machine-readable credibility, and visibility alone will not be enough. The brands that win AI-mediated discovery will be the ones that give machines something worth citing, and buyers something worth trusting."
Shaheen Yazdani · Co-CEO, Intercept
Talk to us
Want us to walk this brief through with your team?
Set up a working session on how the operating-model thesis in this brief applies to your team, budget, and buying committee.
Chapter 4
What to do about it.
The deep dives, the four leadership questions to bring to the next meeting on the calendar, and the closing thesis on why the system is now the advantage.
Sections 11, 12, 13Underneath the radar.
Six deeper looks at the data points that shape the six-pillar view. Open the ones you want.
Agents · Arriving Before The Rules
14% configured prebuilt agents. The rest still debating what an agent is.
14% configured prebuilt agents. The rest still debating what an agent is.
Asked about their grasp of AI agents, the largest group of marketers (14%) report they have configured prebuilt agents rather than built their own. The rest split fairly evenly between not knowing who is supposed to build an agent, assuming it has to be IT, calling everything an agent, and admitting they do not really understand what one is. Most teams meet agents through a vendor demo before they have agreed internally on what one is, who can deploy one, or what an agent is allowed to do. A meaningful number have switched on agents they cannot yet govern.
Frictions · Why Progress Fragments
Top pain points: prioritizing AI use cases (12%), strategic clarity (8%).
Top pain points: prioritizing AI use cases (12%), strategic clarity (8%).
All the top frictions are organizational rather than technical. The missing piece is less another tool than a clearer system for how AI use cases are chosen, owned, and run. The first move is narrowing to a few high-value workflows and naming who owns the agents before switching them on.
Priorities · What Marketers Expect AI To Improve
Content quality 17% · Differentiation 17% · ROI 16% · Production 12% · Lead quality 9% · GEO 7%.
Content quality 17% · Differentiation 17% · ROI 16% · Production 12% · Lead quality 9% · GEO 7%.
Judgment has shifted from whether teams are using AI to what AI produced. The case for any new AI investment now has to lead with the business pressure it answers rather than the features it ships with, and every request that reaches leadership needs to tie to a quality number, a differentiation move, or a revenue growth strategy.
Concerns · Where Trust Breaks First
Attribution 14%. Integration, procurement, automation errors, rogue agents all at 11–12%.
Attribution 14%. Integration, procurement, automation errors, rogue agents all at 11–12%.
The fear has moved from AI itself to losing control of the AI already running inside the function. Attribution standards must be set before AI gets near any workflow that matters, and the line between what an agent can do on its own and where it has to wait for a human has to be drawn before the agent is switched on.
Role Shift · The Marketer Becomes The System Manager
Managing AI agents leads role-change expectations at 18%.
Managing AI agents leads role-change expectations at 18%.
Strategic planning, specialized roles, and technical roles each follow at 10%. The job is shifting from producing the work to judging, supervising, and signing off on it, and the people being asked to do that are the same ones already managing the agents.
Team & Strategy Shift · How The Work Reorganizes
The signals describe restructuring rather than expansion. 59% of committees will demand more proof.
The signals describe restructuring rather than expansion. 59% of committees will demand more proof.
The headcount signals describe restructuring: shifting budget into AI (19%), adding AI specialist roles (18%), reducing execution headcount (15%). Almost no one expects to simply add people. Meanwhile the buying committee itself is changing. As AI tools move into procurement, IT and security weigh in earlier, and 59% of marketers expect their committee to demand more proof before signing off.
Four questions that cannot be deferred.
Each is worth answering before the end of 2026. Bring one to the next leadership conversation on the calendar, and see how aligned the room actually is.
What does "better" actually mean now?
Leadership is asking for quality, differentiation, and ROI. The team needs to know where it is instead still tracking AI by hours saved, and what proof would show AI is improving the work itself, not just the speed of producing it.
What exactly counts as an agent, and who owns its behavior?
The same word now covers assistants, automations, copilots, and systems that act on their own. When one does the wrong thing, somebody carries the consequence, so marketing, IT, legal, and security have to agree on a single answer before an incident decides it for them.
How will we defend AI investments if attribution keeps getting harder?
Attribution is the strongest concern this quarter. The measurement standards must answer, in a defensible and repeatable way, how AI paid for itself when that question lands on the desk.
What would a credible AI operating system look like by year-end?
And who would own how it is governed, measured, and improved? The standardization question is which workflows need to be repeatable enough that any team member can pick one up without depending on a single internal expert. The ownership question is who is building the operating model itself, separate from the people running the daily work inside it.
The system is the advantage.
When execution is abundant and judgment is scarce, the advantage goes to whoever can build the system that makes good judgment repeatable while running the work inside it. That is the partner this moment calls for, one that operates on the system and in the work at the same time.
"It is not an experiment if you know it is going to work. There is no silver bullet for AI, and the only way through is to experiment quickly and take real risks. The job of marketing leadership is to make that safe for their teams."
Andrew Au · Co-CEO, Intercept
Intercept gives B2B tech marketers the operating model. Our team runs the work inside it.
Start the conversation
Give us a hard problem. Let's solve it together.
Take the operating-model thesis from this brief and put it in front of the team that runs it. Tell us where you are and what you are trying to move.
Open the form →