Hybrid Intelligent Enterprise
Hybrid Intelligent Enterprise
hybrid intelligence
hybrid intelligent enterprise
intelligent enterprise
hybrid intelligent teams
AI adoption in enterprise
enterprise AI adoption
how to adopt AI in the enterprise
AI agents and people
human-AI collaboration
AI agents
agentic AI
enterprise AI strategy
AI governance
enterprise memory
earned autonomy
What Is a Hybrid Intelligent Enterprise? A Guide for Leaders
AI agents and people in one team, under one set of rules.
For the last two years, most companies have treated AI as a better tool. They gave people an assistant, tracked how often it was used and counted the hours saved.
That phase is ending. AI agents can now hold a role in a team, work inside the systems the team already uses and pass work back and forth with the people around them.
For leaders, the question has moved on from how much AI the company uses. They now have to decide which roles go to agents, which stay with people, and who answers for each decision.
This guide covers what a Hybrid Intelligent Enterprise is, why it matters now and what it takes to build one. Each section links to a detailed article in this series.
The definition
A Hybrid Intelligent Enterprise is a company where AI agents and people work as one team, each with a job description, under one set of rules.
In this model, AI is a colleague, not a tool. People and agents are both hired into roles, both are accountable for what they do, and both record why they did it. That record stays with the company.
Any role can be held by an AI agent or a person, whether it is doing the work, making the decision or checking it. The mix changes from one process to the next, while the rules stay the same for everyone.
Why now: from assistant to teammate
In most companies today, AI is still an assistant. It drafts when asked and answers when prompted, and a person does the rest. Deloitte's State of AI in the Enterprise 2026 study, based on 3,235 business and IT leaders in 24 countries, found that only 23% of organisations use agentic AI even moderately. Most are still using AI to improve existing processes rather than to rethink how the business runs.
That is changing quickly. Gartner expects at least 15% of day to day work decisions to be made by AI agents on their own by 2028, up from none in 2024. It also expects a third of enterprise software applications to include agentic AI by 2028, up from less than 1% in 2024. Agents are arriving inside the software companies already pay for, as well as in the tools their own teams build.
Leaders expect this shift. In Deloitte's August 2026 survey of 501 US leaders, all at least piloting AI agents, 74% expect nearly half of their business processes to be redesigned around agents within four years. Yet only 15% have scaled agents across functions, and only 5% say their business processes are highly prepared. Most are adding agents on top of the way work already runs.
The same survey shows where the value lies. 75% of leaders say people and AI agents working together create more value than automation alone. Fewer than half, though, have defined how that collaboration works: which decisions agents can make, when people step in, who is accountable and how work is handed over.
Skipping that work is expensive. Gartner expects over 40% of agentic AI projects to be cancelled by the end of 2027 because of rising costs, unclear business value or weak risk controls. Getting it right pays. A Gartner survey of 1,973 managers found that business units which redesign how work gets done, rather than only rolling out AI, are twice as likely to beat their revenue goals. (Detailed article: From AI Assistant to AI Teammate.)
The building blocks
A Hybrid Intelligent Enterprise rests on eight building blocks. Well run companies already manage people this way. The difference is that the same rules now apply to AI agents, in the same way and in the same place.
- Roles for people and AI agents. Every member of the team, person or agent, has a job description: what it does, what it may decide, when it hands off, when it escalates and where its limits sit. An agent is hired into a role the same way a person is. (Detailed article: Hiring AI Into a Role.)
- Decision rights. Every decision has one named owner, AI or person, and never "the system". Nobody, AI or person, checks their own work, so the maker and the checker are always different. (Detailed article: Who Decides, Who Does the Work, Who Checks.)
- Earned autonomy. At first, a person or a second agent checks every task an agent completes. As the record shows the agent getting things right, its authority widens one task at a time, and when the record shows mistakes, it narrows again. The agent's own track record decides how far its authority goes. (Detailed article: Earned Autonomy.)
- Work that moves both ways. Work passes from agents to people and back again, as it would in any good team. An agent might draft a reply that a person improves, or prepare an order that a person approves. High-risk or complex calls escalate to a named person, then to a manager, then to leadership. (Detailed article: Work That Moves Both Ways.)
- Governance inside every task. A policy document is where governance starts. In a Hybrid Intelligent Enterprise, the rules run inside every task and every decision, every day, for people and agents alike. (Detailed article: Governance That Runs Inside Every Task.)
- Access and identity. Every agent has its own login and only the access its role needs: the systems it can use, what it can do in each and the data it can see. It never borrows the access of the person who set it up. (Detailed article: Access and Identity for AI Agents.)
- Cost control. The cost of each task is estimated and capped before the work runs, and charged to a business budget, so leaders see AI spend before the invoice arrives. (Detailed article: Knowing the Cost Before the Work Runs.)
- Enterprise memory. Every task, decision and handoff is recorded with its reason, from AI and people alike, and kept on the company's side. Over time, this becomes the company's memory of how it decides. It grows with every decision and stays when people leave or vendors change. (Detailed article: Enterprise Memory.)
One principle holds the eight together. Built agents, agents bought from SaaS or AI agent vendors, and people all follow one set of rules that the company owns. (Detailed articles: Bought, Built and People: One Set of Rules, and Measuring a Hybrid Intelligent Enterprise.)
What it looks like on LeafCraft
LeafCraft already runs this way of working for companies in consumer electronics, AdTech and B2B SaaS. Their names are kept private here, but the work, the people involved and the results are real. Each story names the industry, the process and the use case, then explains what was going wrong, how agents and people now share the work, and what changed.
1. Consumer electronics: one AI across a phone and a pair of earbuds
Industry: consumer electronics. Process: product experience. Use case: one set of finance, health, career and lifestyle agents serving voice requests on AI earbuds and touch requests on a smartphone.
An Indian consumer electronics brand sells both smartphones and AI earbuds, and wanted AI built into both, so the devices could help people every day with their finances, health, career and lifestyle. That raised a hard design problem. People talk to their earbuds and tap on their phones, so the same help had to work through two very different inputs. Done the usual way, each device would have needed its own AI stack. A request started by voice on the earbuds would lose its context the moment the user picked up the phone. Voice commands could trigger actions with no check at all. And the brand's customers speak many Indian languages as well as English.
What LeafCraft did. LeafCraft gave the brand one platform for both devices. Here is how it helped under each of its five pillars.
- AI and people as one team: LeafCraft's team built 10 agents for the brand, covering finance, health, career and lifestyle, and gave each one a clear role. When a user asks for something, by voice on the earbuds or in the app, LeafCraft sends the request to the agent whose job it is. Where an action needs a check, the agent waits for confirmation before it acts, so the user stays in charge of anything that changes their device.
- Governance: Every agent has a job description and its own level of autonomy, set by the brand. Safe actions that are easy to undo can run on their own, and anything that changes the device or the user's data must pass a check first. This closed the risk of a voice command acting with no check at all.
- Cost control: The cost of every request is known and capped before it runs, on both devices. The brand can grow usage without its AI bill growing out of sight.
- One view: All 10 agents sit on one roster that serves both the earbuds and the phone. The brand runs one set of agents and one set of rules, instead of building and maintaining a separate AI stack for each device.
- Enterprise memory: Both devices share one memory. A conversation started on the earbuds carries on in the app, so the user never has to repeat themselves, in English or in any of the Indian languages the agents support.
How a typical request moves through the team.
- What the agent did: On a morning walk, a user asks the earbuds, in Hindi, how much they spent on food delivery last month. The finance agent answers. The user then asks it to send that summary to a family member.
- Why it went to a person: Sharing personal data is an action that must pass a check, so the agent does not send it on its own. It asks the user to confirm.
- What the person did: When the user picks up the phone, the app shows the summary and the contact. The user removes one line they would rather keep private and confirms.
- How it was resolved: The agent sends the edited summary and confirms on the earbuds that it has gone.
- What LeafCraft recorded: The request, the agent that handled it, why the check was needed, what the user changed and the final action are all recorded on LeafCraft. Over time, this shows the brand which actions users change most before they confirm, so it can tune each agent's role and limits from real evidence. That record is the brand's enterprise memory.
The result is 10 governed agents on one roster, helping users with finance, health, career and lifestyle, working in English and most Indian languages by voice or app, with one memory across both devices. The brand did not have to build and maintain two separate AI systems.
2. Consumer electronics: turning support into a reason to use the phone
Industry: consumer electronics. Process: smartphone customer support. Use case: on device agents that help customers before something goes wrong.
The same brand had a familiar problem with support. Every phone it sold added to the support bill, and help only started once something had gone wrong. Support did nothing to make people use the phone more, at a time when a young phone brand badly needed to stand out. On top of that, customers' personal data had to stay on the device.
What LeafCraft did. LeafCraft's team built 9 agents that live on the phone. Here is how the platform helped under each of its five pillars.
- AI and people as one team: The agents answer common support questions and offer help before anyone asks for it. When a question needs a person, the agent hands it to the support team with the case recorded in Zoho, so the person starts with the whole story instead of starting from scratch.
- Governance: Every agent has a job description and a named owner. Any high risk action on the device, such as one that changes settings or data, always goes to a person first. As an agent's record shows it handles a type of request well, it is allowed to resolve more of those requests on its own.
- Cost control: The cost of every support interaction is known and capped before the agent starts work. This is how the brand brought the cost of each interaction down, instead of watching support costs grow with every phone sold.
- One view: The agents and the people who take escalations sit on one roster, so the support head can see who resolved what, and where people are still needed.
- Enterprise memory: Every case and how it was resolved is recorded, so the next similar case starts from what the brand already knows. Personal data never leaves the phone, so this memory does not come at the cost of customer privacy.
How a typical case moves through the team.
- What the agent did: A customer says their battery is empty by lunchtime. The support agent checks battery use on the phone, finds one app running in the background and offers to restrict it. The customer agrees, and the agent changes the setting.
- Why it went to a person: Two days later the drain is still there. The software fix has not worked, and a possible battery fault is outside the agent's role, so it hands the case to the support team in Zoho with everything it has already tried.
- What the person did: A support executive reads the case, sees the battery health data and the steps already taken, and books a battery check at the nearest service centre.
- How it was resolved: The service centre replaces the battery under warranty. A week later, the agent checks in with the customer to confirm the phone now lasts the day.
- What LeafCraft recorded: The agent's diagnosis, why it handed over, what the person decided and how the case ended are all recorded on LeafCraft, while personal data stays on the phone. When the same pattern shows up on another phone, the agent recognises it sooner and brings in a person straight away. Every case adds to the brand's enterprise memory.
Phone and feature usage rose by about 45%. About 30% of support queries are now resolved without a person, and the cost of each support interaction fell by about 25%. Support stopped being only a cost and became a reason for customers to use the phone more.
3. Consumer electronics: running operations with agents and people together
Industry: consumer electronics. Process: business operations across ERP, HR and marketing. Use case: agents prepare purchase orders, match invoices and onboard employees, and people approve.
Inside the business, the same brand's AI efforts had grown team by team, with no one in control of the whole. Work moved by hand between SAP, Zoho and spreadsheets. Approvals were chased over email and left no record. Much of how processes really worked sat in people's heads rather than with the company, and nobody could forecast or cap what AI was costing.
What LeafCraft did. Here the brand's own team builds the agents on LeafCraft, using the LeafCraft SDK and Studio, with guidance from LeafCraft's team. Here is how the platform helped under each of its five pillars.
- AI and people as one team: Agents take the routine work: preparing purchase orders in SAP, matching invoices, and handling employee onboarding through Zoho, Google Sheets and the brand's own app. Big calls go to named owners, who approve the work or send it back with a reason, and the agent picks it up again from there.
- Governance: Every agent has a job description, a named owner and clear approval rules, and no agent approves its own work. Approvals now happen inside the task, so nobody chases them over email.
- Cost control: Every agent has a cost cap, and the cost of each task is known before it runs. For the first time, the brand can forecast and limit what AI is costing it.
- One view: AI work that had grown team by team now sits on one roster, alongside the people in finance, HR and marketing who work with the agents. Leaders can see every agent, what it does and who owns it.
- Enterprise memory: Every decision and the reason behind it is recorded and stays with the brand, so process knowledge no longer lives only in people's heads. As an agent's record shows it getting things right, its role widens.
How a typical case moves through the team.
- What the agent did: The invoice matching agent compares a supplier's invoice in SAP with the purchase order and the goods receipt. The quantities match, but the unit price is higher than the price on the purchase order.
- Why it went to a person: Its job description allows it to clear only invoices that match. A price difference is a call for a person, so it sends the invoice to the named owner in procurement, with the purchase order, the receipt and the difference highlighted.
- What the person did: The procurement manager checks and finds that the supplier raised prices after a rise in raw material costs. The new price was agreed by email but never updated on the purchase order. She approves the invoice and asks for the purchase order to be corrected.
- How it was resolved: The agent updates the purchase order in SAP, clears the invoice for payment and lets the supplier know.
- What LeafCraft recorded: The mismatch, the manager's decision and her reason, the agreed price rise, are recorded on LeafCraft with the invoice. The next time that supplier bills at the new price, the agent has the answer on record and does not need to ask again. The decision no longer sits in an email thread. It is part of the brand's enterprise memory.
Across ERP, HR and marketing operations, the brand reports about a 30% gain in operational efficiency. Invoice matching takes about 25% less work, agents now handle about 30% of routine HR requests, and marketing launches go out about 20% faster.
4. AdTech: an AI DataOps team for ad operations
Industry: AdTech. Process: ad operations and yield management. Use case: an AI DataOps team that turns raw ad data into yield actions.
An AdTech data and operations company had analysts spending 40 to 60 hours a week on manual data work. By the time the numbers were ready, the window to act had often passed, so yield problems leaked revenue every day. Tools were spread across many data sources with no single control, and the reasoning behind decisions lived with individual people.
What LeafCraft did. The company now runs an AI DataOps team on LeafCraft. Here is how the platform helped under each of its five pillars.
- AI and people as one team: Agents read ad and revenue data from the ad server and Google BigQuery, spot yield problems and recommend what to do. The ad ops team approves those actions in Slack, so nobody has to learn a new portal. Exceptions go to analysts, and their decisions go back to the agents, so the next similar case is handled the same way.
- Governance: Every agent works under clear rules, and every action it takes on the ad server is audited. Low risk actions are handed to the agents as trust in their record grows, while everything else waits for a person's approval.
- Cost control: Every agent's cost is capped before it runs, so agents can work across every data source, every day, within a known budget.
- One view: Work that was spread across many tools and data sources now runs on one platform, under one control. Every action is tracked in Jira until it closes, so the team always knows what is open and who owns it.
- Enterprise memory: Every decision is kept with its reason, so the logic behind yield calls belongs to the company rather than to individual analysts.
How a typical case moves through the team.
- What the agent did: Overnight, an agent notices that fill rate on a premium ad unit has dropped sharply after a floor price change. Using data from the ad server and BigQuery, it works out the likely cause and recommends lowering the floor for that unit.
- Why it went to a person: Changing a floor price affects revenue and is outside the low risk actions the agent may take alone, so it posts the recommendation in Slack for the ad ops team, with the data behind it.
- What the person did: An ad ops lead agrees with the cause but chooses a smaller cut, because a large advertiser's campaign starts that week. She approves the adjusted change in Slack.
- How it was resolved: The agent applies the approved floor on the ad server, updates the Jira ticket and closes it once fill rate recovers.
- What LeafCraft recorded: The drop, the agent's recommendation, the lead's change and her reason, the action on the ad server and the outcome are all recorded on LeafCraft. The next time a floor change hits fill rate during a big campaign, the agent recommends with that context. The reasoning stays with the company as enterprise memory, even if the lead moves on.
Analysts now do about 50 to 60% less manual data work. The company recovers more than $12,000 a day in revenue, saves about $200,000 a year, and went from its first workshop to production in six weeks.
5. B2B SaaS: an agent product built on LeafCraft
Industry: B2B SaaS. Process: customer success. Use case: an agent product that runs customer success follow ups and books meetings end to end.
A B2B SaaS company wanted to sell a product in which agents coordinate customer success for its own customers. Building the orchestration in house would have delayed launch by months. Governance and handoffs between agents and people would have to be built from scratch, every new use case would take months of engineering, and without a predictable cost per agent there was no sound way to price the product.
What LeafCraft did. The company built its whole agent product on LeafCraft, instead of building orchestration and governance itself. Here is how the platform helped under each of its five pillars.
- AI and people as one team: More than 12 agents run follow ups and book meetings end to end through the product's CRM and calendars, using account signals from Clay and LinkedIn and tool integrations through Composio. When an account needs a human touch, the agent hands it to a customer success manager in Slack.
- Governance: Rules and handoffs between agents and people come built in, so the company did not have to build them from scratch. New use cases are set up through configuration rather than months of new code.
- Cost control: Every agent's cost is capped, which gives the company a predictable cost per agent to price its own product on.
- One view: All of the agents run on one platform, so the company sees and manages every agent in production in one place.
- Enterprise memory: Agents and customer success managers share the same account history, so whoever picks up an account, agent or person, knows what has already happened.
How a typical case moves through the team.
- What the agent did: The follow up agent sees from the CRM and LinkedIn signals that the main contact at a customer account has changed jobs, and that product usage has dropped over the past month. It drafts an email to the new contact and proposes a meeting time from the calendar.
- Why it went to a person: The account is up for renewal soon, and the drop puts it at risk. Accounts at risk are handed to a customer success manager, so the agent passes it over in Slack with the account history and its draft.
- What the person did: The customer success manager rewrites the email to mention a feature the customer asked for last quarter, and asks the agent to set up a call with both the new contact and the account's executive sponsor.
- How it was resolved: The agent books the meeting, sends reminders and, after the call, the manager records that the renewal is back on track.
- What LeafCraft recorded: The signal, the agent's draft, the manager's change and why she made it, and the outcome are all kept in the shared account history on LeafCraft. Whoever touches the account next, agent or person, sees the full story, and the company builds a record of which signals point to risk. That is its enterprise memory.
The company launches a new use case about 30% faster than it could by building in house, runs more than 12 agents in production, and spends about 30% less engineering effort on agent infrastructure.
Figures are as reported by each customer.
What a Hybrid Intelligent Enterprise is not
It is sometimes treated as a maturity score. One score can't describe a company that makes thousands of different decisions a day, so it is better judged one decision at a time: who leads, who does the work and who checks.
The number of AI tools a company runs doesn't measure it either. Agents with no named owner add up to sprawl.
It doesn't mean AI replacing people. AI and people sit on the same org chart in the same kinds of roles, and Gartner expects less than 1% of US jobs to be lost to AI through 2028.
No vendor can hand it to you. A vendor governs the agents inside its own product, while your rules, your cost limits and the record of your decisions need to stay with you.
What it means for each leader
CEO and managing director. AI pays back when the work itself is redesigned. Your part is choosing which processes go first and making the business own the result alongside IT.
CIO, CDAO and Chief AI Officer. Agents now arrive from SaaS vendors and from your own teams. One view of all of them, with the rules held on your side, is what lets you move from pilots to scale without losing control.
CFO and finance leaders. AI work should have a known cost before it runs, and a cap. Finance also needs maker and checker kept apart, and a record of every decision and its reason that auditors can open.
COO, supply chain and procurement. An agent that changes an order or a supplier commitment needs clear limits, a clean handoff and a fast route to a named person when something falls outside the rules.
Commercial, sales and revenue growth leaders. Pricing, schemes, claims and customer terms carry real money. Each of these decisions needs one named owner, and an agent's authority should grow only as its record earns it.
Risk, audit and compliance. After any incident, the first question is who approved it and why. The answer should come from a record kept on the company's side that nobody can change later.
CISO. Every agent is an identity. It needs its own login, the narrowest access its role allows, and that access removed the day the agent is retired.
Product and technology leaders at B2B software companies. Your customers will ask how the agents in your product are governed. Agents that fit inside a customer's own rules and record are becoming part of what you sell.
How it works in real teams
Here are six real teams. In each one, every role has an owner, and that owner is either an AI agent or a person.
A software development team, shipping a change to the checkout page
| Role | Held by | What they do |
|---|---|---|
| Product manager | Person | Writes the story in Jira and sets what done means |
| Coding agent | AI agent | Writes the change and opens a pull request in GitHub |
| Test agent | AI agent | Runs the regression suite in the CI pipeline and flags a failing payment test |
| Senior developer | Person | Reviews the pull request, fixes the edge case the coding agent missed and approves the merge. The coding agent never approves its own work |
| Release agent | AI agent | Drafts the release notes and checks that every change in the release has a passing test |
| On call engineer | Person | Decides when the change goes to production |
What a Hybrid Intelligent platform does here:
- AI and people as one team: The coding agent writes the change, a developer reviews it, the test agent checks it and an engineer decides when it ships. The work passes back and forth inside the same pull request.
- Governance: The coding, test and release agents each have a job description and a named owner, and the coding agent can never approve its own pull request.
- Cost control: The cost of each agent's run is known and capped before it starts, so a long test cycle cannot quietly eat the month's budget.
- One view: The engineering head sees every agent, whether the team built it, bought it with a tool or built it on the platform, on one roster with the developers.
- Enterprise memory: Why the senior developer changed the code, and why the on call engineer picked that release window, is recorded and stays with the company.
LeafCraft gives engineering teams this, with every agent and developer under one set of rules.
An FMCG sales team, planning tomorrow's orders for a town
| Role | Held by | What they do |
|---|---|---|
| Sales planning agent | AI agent | Reads overnight secondary sales in the distributor management system, finds outlets that have not reordered a fast moving pack and builds the beat plan and suggested orders in the sales force app |
| Salesman | Person | Visits each outlet, changes quantities based on the shelf and what the retailer can pay, and books the order |
| Area sales manager | Person | Approves an extra scheme for a slow outlet |
| Scheme check agent | AI agent | Checks the scheme against the budget and scheme rules before it goes live, and records why it was given |
| Billing agent | AI agent | Raises the invoice in the distributor's system and sends any order above the retailer's credit limit back to the salesman |
What a Hybrid Intelligent platform does here:
- AI and people as one team: The agent plans the beat overnight, the salesman corrects it at the shelf and the billing agent picks up his changes. Each one improves the other's work on the same order.
- Governance: The scheme rule, with its budget and the manager allowed to approve it, is written once, and the scheme check agent applies it before anything goes live.
- Cost control: The cost of building the nightly plan for every beat is known and capped before it runs.
- One view: The sales head sees the three agents, the salesmen and the area sales managers on one roster, and who did what on each beat.
- Enterprise memory: Every override is recorded with its reason, such as a salesman cutting an order because the retailer was short of cash, so the next plan can use what happened at the shelf.
LeafCraft gives sales teams one roster of agents and people, with schemes, budgets and reasons in one place.
A consumer durables service team, fixing an air conditioner that is not cooling
| Role | Held by | What they do |
|---|---|---|
| Service desk agent | AI agent | Answers the customer on WhatsApp, checks the warranty in the CRM, looks up common faults for that model and books a technician |
| Field technician | Person | Diagnoses the unit on site, finds a compressor fault and raises a replacement request |
| Warranty agent | AI agent | Checks the claim against the warranty terms and the part's history. Approves claims within its limit and sends larger ones to the service manager |
| Service manager | Person | Approves the compressor replacement |
| Parts agent | AI agent | Raises the part order in SAP, keeps the customer updated and closes the job once the customer confirms the fix |
What a Hybrid Intelligent platform does here:
- AI and people as one team: The agents handle the customer and the paperwork, the technician handles the machine and the manager handles the exception. The job moves between them and the customer never has to repeat the problem.
- Governance: The warranty agent's approval limit is set by its owner. Above that limit, the claim goes to the service manager every time.
- Cost control: The cost of every service job is known and capped before the agents start work on it.
- One view: The service desk agent that came with the CRM and the warranty agent built in house sit on one roster with the technicians and the service manager.
- Enterprise memory: The fault, the decision and the reason are recorded, so the company builds its own history of which models fail and how they were fixed.
LeafCraft lets service teams run bought and built agents side by side, under one set of rules.
A bank's SME lending team, deciding a working capital loan
| Role | Held by | What they do |
|---|---|---|
| Credit analysis agent | AI agent | Collects GST returns and bank statements, runs the bureau check and drafts the credit memo in the loan origination system |
| Relationship manager | Person | Visits the factory, adds what the numbers do not show and owns the customer relationship |
| Credit officer | Person | Makes the lending decision |
| Policy check agent | AI agent | Checks the decision against the bank's lending policy and limits before the sanction letter goes out, and sends any exception back to the credit officer |
| Documentation agent | AI agent | Prepares the sanction letter and keeps the full trail ready for the auditors |
What a Hybrid Intelligent platform does here:
- AI and people as one team: The agent gathers the papers and drafts the memo, the relationship manager adds what she saw at the factory, the credit officer decides and another agent checks that decision. The file moves from agent to person and back until the loan is sanctioned.
- Governance: Each agent has a job description, a named owner in the credit team and clear rules on what it can see and do. The agent that drafts the memo never checks the decision.
- Cost control: The cost of preparing each credit memo is known and capped before the work starts.
- One view: The head of credit sees every agent and every person working on loan files on one roster, with who holds each step.
- Enterprise memory: The decision, the reasons and every approval are recorded, which gives auditors a full trail and gives the bank a memory of why it lent.
LeafCraft gives lending teams agents with clear owners and a decision trail auditors can follow.
An insurer's motor claims team, settling a claim for a damaged car
| Role | Held by | What they do |
|---|---|---|
| Claims intake agent | AI agent | Takes the photos and documents the policyholder uploads in the app and checks the policy cover |
| Assessment agent | AI agent | Compares the garage estimate with the photos and settles small claims within its limit |
| Surveyor | Person | Inspects the car when the claim is above that limit |
| Claims manager | Person | Approves large claims and any claim the agents flag |
| Fraud check agent | AI agent | Before any payment, including the ones the assessment agent settles, compares the claim with past claims and flags a garage that keeps appearing in similar cases |
What a Hybrid Intelligent platform does here:
- AI and people as one team: Agents settle the simple claims, the surveyor and the claims manager take the hard ones, and an agent checks every payment either way. The policyholder sees one claim handled by one team.
- Governance: The assessment agent's settlement limit is set by the claims manager who owns it, and the fraud check agent reviews every payment, including the ones another agent approved.
- Cost control: The cost of handling each claim is known and capped before it is processed.
- One view: Claims leaders see the agents, surveyors and claims handlers on one roster.
- Enterprise memory: Each flag and its outcome is recorded, so the pattern behind a suspicious garage stays with the insurer.
LeafCraft lets claims teams give agents real limits, with every payment checked and recorded.
A manufacturing plant's maintenance team, acting on rising vibration on a bottling line
| Role | Held by | What they do |
|---|---|---|
| Condition monitoring agent | AI agent | Watches the sensor data, spots the rising vibration and pulls the machine's history from SAP Plant Maintenance |
| Maintenance planning agent | AI agent | Checks the production plan in the MES, drafts the work order, suggests the next planned stop and reserves the spare part from stores |
| Maintenance head | Person | Approves the work order and agrees a Sunday stop with the production planner |
| Production planner | Person | Moves the affected batch to another line |
| Technician | Person | Carries out the repair |
| Quality engineer | Person | Signs off the first bottles off the line before full production restarts |
What a Hybrid Intelligent platform does here:
- AI and people as one team: Agents spot the problem and plan the fix, people agree the stop and carry out the repair, and the quality engineer signs off before the line restarts.
- Governance: The agents can draft the work order and reserve parts, but stopping the line is a person's call, and that approval step always applies.
- Cost control: The cost of each agent task is known and capped before it runs.
- One view: The plant head sees the agent that came with the monitoring system and the agent built in house on one roster with the maintenance head, the planner, the technician and the quality engineer.
- Enterprise memory: Why the stop moved to Sunday, what was found and how it was fixed are recorded, so the next team facing the same machine starts from the plant's own history.
LeafCraft lets plants run agents next to their people, with stopping the line always a person's call.
The pattern is the same in every team. Each role has a job description and an owner. Agents and people hold roles side by side, and work moves between them in both directions. No one checks their own work, so sometimes an agent checks a person and sometimes a person checks an agent. LeafCraft is the platform that holds this together: the roles, the rules, the budget and the memory, for agents built or bought and for the people who work with them. People and AI as one team. You own the rules, memory and budget.
Where to start
There is no need to redesign the whole company at once. Start with one process.
- Name one senior owner for the rules, with the CIO, the business head and risk at the table.
- Pick one process and mark every role in it as a person or an AI agent, built or bought.
- Give every role a job description, a checker and an escalation path, and give every decision one named owner.
- Set access for each role, starting narrow.
- Set cost limits per role and per task before any work runs.
- Start the record on day one, with every task, decision and handoff and the reasons of AI and people.
- Widen authority when the record earns it, and narrow it when the record shows mistakes.
- Review the rules against the record on a fixed schedule.
(Detailed article: Where to Start: One Process, Step by Step.)
How LeafCraft helps
LeafCraft is the platform for the Hybrid Intelligent Enterprise, where AI agents and people work as one team, each with a job description, under one set of rules. Enterprises run the agents they've built or bought (from SaaS or AI agent vendors), or build new ones on LeafCraft. It works with the workflow and automation tools they already use, whether built in house, bought as software or custom developed.
It rests on four pillars:
- Governance: every agent and person has a job description, a named owner, clear rules and approvals, and no one checks their own work.
- Cost control: the cost of every task is known and capped before it runs.
- One view: enterprises see every agent, built or bought, alongside their people on one roster.
- Enterprise memory: every decision and its reason is recorded and stays with the enterprise.
People and AI as one team. You own the rules, memory and budget.
If you are planning the move from AI tools to AI teammates, we would be glad to show you how it works. Book a conversation
Frequently asked questions
What is a Hybrid Intelligent Enterprise?
A Hybrid Intelligent Enterprise is a company where AI agents and people work as one team. Each agent and each person has a job description, a named owner and clear rules. Any role in a process, whether doing the work, deciding or checking, can be held by an agent or a person, and work moves between them in both directions under one set of rules.
How is a Hybrid Intelligent Enterprise different from giving everyone an AI assistant?
An AI assistant waits to be asked and helps one person work faster. In a Hybrid Intelligent Enterprise, an agent owns part of the work inside a team. It has a role, an owner and limits, it hands work to people when a decision needs them, and people hand work back to it.
What are Hybrid Intelligent Teams?
Hybrid Intelligent Teams are teams in which AI agents and people hold roles side by side in the same process. In a bank's lending team, for example, an agent drafts the credit memo, the relationship manager adds what she saw at the customer's factory, the credit officer decides, and a separate agent checks that decision against lending policy.
Why does this matter now?
Agents can now do real work inside the systems teams already use, but most companies are adding them to old ways of working. Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, and Deloitte found that most leaders are layering agents onto existing processes instead of redesigning how work gets done. Redesigning the team is what turns agents into results.
What are the building blocks of a Hybrid Intelligent Enterprise?
There are eight: roles for agents and people, clear rights over who does, decides and checks, earned autonomy, work that moves both ways, governance inside every task, access and identity for agents, cost control before work runs, and enterprise memory.
Can an AI agent check a person's work?
Yes. The rule is that no one checks their own work. Sometimes a person checks an agent, such as a senior developer approving a coding agent's pull request. Sometimes an agent checks a person, such as a policy agent reviewing a credit officer's decision before the sanction letter goes out.
What is earned autonomy for AI agents?
Earned autonomy means an agent starts with narrow limits and a person approving its work. As its record shows it getting things right, it is trusted with more. How much an agent may do on its own grows from evidence, and it can be narrowed again if the record changes.
What is enterprise memory?
Enterprise memory is the record of every decision made by agents and people, along with the reason behind it, kept by the company. It means knowledge stays when people move on, and agents can learn from how people resolved similar cases before.
How do you control the cost of AI agents?
Know and cap the cost of every task before it runs, agent by agent and process by process. That lets finance forecast AI spend and stops a single long running task from using up a month's budget.
Can agents bought from SaaS vendors and agents built in house work under the same rules?
They should. Agents that come with SaaS tools, agents bought from AI agent vendors and agents a company builds itself should all sit on one roster, next to the people they work with, under one set of rules for roles, approvals, cost and memory.
Where should a company start?
Start with one process that matters. Map who does, decides and checks each step, choose which roles go to agents, set the rules and approvals, cap the cost, and record every decision. Widen the agents' roles as their record builds, then move to the next process.
What is LeafCraft?
LeafCraft is the platform for the Hybrid Intelligent Enterprise, where AI agents and people work as one team, each with a job description, under one set of rules. Enterprises run the agents they've built or bought (from SaaS or AI agent vendors), or build new ones on LeafCraft.
LeafCraft rests on five pillars, and each one solves a problem leaders hit when agents move into real work.
- AI and people as one team: Agents and people hold roles side by side in the same process. Work moves between them in both directions, and agents hand a decision to a person whenever it needs one.
- Governance: Every agent and every person has a job description, a named owner, clear rules and approvals, and no one checks their own work.
- Cost control: The cost of every task is known and capped before it runs, so AI spend can be forecast and limited.
- One view: Every agent, built or bought, sits on one roster alongside the people it works with, so leaders can see who is doing what.
- Enterprise memory: Every decision and the reason behind it is recorded and stays with the enterprise, so knowledge stays when people move on.
Together, they let a company run AI agents and people as one team while it keeps ownership of the rules, the memory and the budget.
How does LeafCraft help a company become a Hybrid Intelligent Enterprise?
LeafCraft gives every agent a role on the same team as the people it works with. It holds the rules and approvals for each process, caps the cost of each task before it runs, shows every agent and person on one roster, and records every decision with its reason. Companies can bring agents they have already built or bought, or build new ones on LeafCraft, and widen each agent's role as its record grows.
Which industries use LeafCraft today?
LeafCraft is a horizontal platform. It works as the AI infrastructure for hybrid intelligence, so it can help in any industry and in any process where AI agents and people work together. Industries LeafCraft supports today include:
- FMCG and consumer goods
- Consumer durables
- Electronics and IoT
- Paints and coatings
- Liquor and beverages
- Auto components
- Chemicals and materials
- B2B software
- Banking, NBFCs and insurance
- Global capability centres
- Logistics and supply chain
- Advertising and media
Whatever the industry or process, LeafCraft brings the same five pillars:
- AI and people as one team: Agents and people hold roles side by side in the same process, and work moves between them in both directions.
- Governance: Every agent and every person has a job description, a named owner, clear rules and approvals, and no one checks their own work.
- Cost control: The cost of every task is known and capped before it runs, so AI spend can be forecast and limited.
- One view: Every agent, built or bought, sits on one roster alongside the people it works with.
- Enterprise memory: Every decision and the reason behind it is recorded and stays with the enterprise.
Sources
All figures were checked against the publisher's own page or a report on it. All are from 2025 or 2026.
- Deloitte Insights, "AI agents are only the beginning: The path to agentic transformation", 12 Aug 2026. Survey of 501 US leaders, all at least piloting agentic AI: 74% expect nearly half of processes redesigned within four years; 15% have scaled multi-agent adoption; 5% say processes are highly prepared; 75% see more value in human and agent collaboration; fewer than half have defined human and agent operating models.
- Deloitte press release, 12 Aug 2026. Most organisations are adding agents on top of existing processes.
- Unite.AI on Deloitte, "State of AI in the Enterprise 2026". 3,235 leaders in 24 countries; 23% use agentic AI at least moderately; most still use AI to optimise existing processes.
- Gartner press release, 25 Jun 2025. Over 40% of agentic AI projects cancelled by end of 2027; 15% of day to day work decisions made autonomously by 2028; 33% of enterprise software applications include agentic AI by 2028.
- Gartner press release, 27 Oct 2025. Survey of 1,973 managers: units that redesign work are twice as likely to beat revenue goals; less than 1% of US jobs lost to AI through 2028.
- Customer examples: LeafCraft customer results, anonymised and as reported by each customer.