Entrepreneur Greg Isenberg recently spent 48 hours in San Francisco with AI founders and executives running companies valued between $1 billion and $100 billion. He came home with 21 observations about where the AI industry is headed. Most of them concern venture funding and startup strategy. A handful apply directly to anyone responsible for a fleet.

That overlap matters because AI is reaching fleet operations through tools you already pay for. Telematics platforms, maintenance software, fuel card portals, and leasing dashboards are all adding AI features. Some are adding agents, which are software programs that take action on their own instead of waiting for someone to click a button.

The fleet management trends worth watching in 2027 center on readiness. Can your data, your processes, and your vendors support software that acts without a person driving every step? Below are six shifts from Isenberg's list, translated for fleet managers and for the operations, HR, finance, and procurement leaders who share responsibility for company vehicles.

1. Fleet Data Cleanup Comes Before AI

Isenberg's most practical observation came from companies trying to connect their internal data to AI agents. Nearly every executive he spoke with wanted to pull company data into one place so agents could use it. Then they looked closely and found a mess: the same customer listed five different ways, records scattered across old chat channels, and PDFs no one could search.

Fleet programs have this problem in a concentrated form. A single vehicle's story is usually spread across systems. Telematics holds location and odometer readings. The fuel card provider holds transactions. Maintenance records sit with shops, in a software platform, or in a folder of scanned invoices. Lease terms live in the lessor's portal. Toll charges arrive on a separate statement. Driver assignments live in HR, and at least one spreadsheet ties it all together because none of the systems talk to each other.

Each system also tends to identify the vehicle differently. One uses the VIN, another uses a unit number, a third uses the license plate, and the fuel card might be tied to a driver's name.

People work around these gaps because they know the history. An AI agent has no such context. Ask an agent which vehicles should be replaced next year, and it will pull odometer data from telematics, repair costs from maintenance records, and lease end dates from the lessor. If those records disagree, the agent will still produce an answer. It will simply be wrong, and it will sound confident.

The fleets that get real value from AI in 2027 will be the ones that do unglamorous data work first:

• Pick one master identifier. The VIN is the natural choice, and every other system should map back to it.
• Reconcile the active vehicle list quarterly. Compare telematics, fuel cards, insurance schedules, and lease records, and resolve every mismatch.
• Assign ownership for driver-to-vehicle records. Decide whether HR or fleet owns assignments and how changes flow to fuel cards and telematics.
• Digitize paper and PDF records. Inspection reports, titles, and registrations locked in scanned files are invisible to software.
• Retire shadow spreadsheets. Every extra copy of the vehicle list becomes another version of the truth.

This work pays off even if you never adopt an AI tool. Clean records make telematics data analytics more reliable, simplify audits, and shorten the time it takes to answer a leadership question about fleet cost.

2. Screen-Heavy Fleet Software Will Feel Its Age

Isenberg predicted that a good share of legacy software will not survive the shift to AI agents, especially tools built around people clicking through screen after screen. Agents work through connections between systems. A platform that requires a person to log in and navigate a dozen pages to finish a task becomes a bottleneck. Think about the routine tasks your team handles each week. Approving a maintenance purchase order, reassigning a vehicle to a new hire, ordering a replacement unit, and updating a driver's contact details can each take several logins across several portals. Those are exactly the tasks agents are good at, provided the software underneath lets them in.

A related point from his list deserves attention. Isenberg argued that as software becomes easier to copy, the lasting value shifts to the data inside it. For a fleet, that means years of maintenance history, fuel transactions, accident records, and replacement decisions. That history is what makes any future analysis or AI tool useful, and it belongs to you.

Three steps help here:

1. List the tasks your team repeats most often and note which systems each one touches.
2. Ask every fleet vendor whether those tasks can run through an API or a native integration.
3. Review contracts for data ownership and export rights, so you can take your full history with you if you switch providers.

Vendors with open integrations and clean data exports will fit into an AI-assisted workflow. The others will slow it down.

3. The AI Model Is the Smallest Part of a Good Fleet Tool

The strongest AI teams Isenberg met treat the underlying model as interchangeable. Several route different jobs to different models and swap them as better ones arrive. What they invest in is everything surrounding the model. He described a useful industry-specific AI product as four components, with the model accounting for maybe a fifth of the value.

That breakdown doubles as a buyer's checklist. When a fleet vendor pitches an AI feature, use the four components below to judge it.

Isenberg noted that the talent AI companies want most is someone with good judgment, because the difficult work now lies in deciding which AI output is good. In fleet terms, a tool that drafts a replacement plan is only as good as the checks confirming it used the right cycle, the right resale assumptions, and the right vehicle class for each job.

Isenberg also mentioned that strong teams judge AI output by what users do with it: whether they edit it, export it, or share it. That is a sensible test for your own pilots. If your team rewrites every recommendation the tool produces, the tool is not saving time yet.

4. Human Checkpoints and Audit Trails Become Standard

One of the more memorable lines Isenberg heard came from a robotics founder, who said the hardest problems in AI today involve deployment more than research. That founder's robots work on their own and call a human when they get stuck. Each of those calls becomes training data that makes the robot better. The question for 2027 is where to draw the lines for AI. A sensible starting framework sorts fleet decisions into three groups:

• Low-risk actions an agent can complete on its own: scheduling preventive maintenance, sending renewal reminders, flagging an expired registration, or matching fuel transactions to vehicles.
• Medium-risk actions that need a quick approval: repairs above a set dollar threshold, vehicle reassignments, and driver coaching alerts before anyone acts on them.
• High-stakes decisions that stay with people: vehicle orders, remarketing timing, policy exceptions, and any disciplinary action tied to driver behavior.

The handoff points matter as much as the automation. Every time a person corrects or overrides the agent, that correction should be captured and used to tighten the rules.

Isenberg also predicted that insurance for AI agents is coming, and that insurers will require audits once an agent makes an expensive mistake. Fleet programs already answer to insurers, auditors, and finance teams, so expect those groups to ask how AI-driven decisions were made. Any tool you adopt should keep a log of what the agent did, which data it used, and who approved the action. If a vendor cannot produce that record, the tool creates more risk than it removes.

5. Every New Integration Opens Another Door

Cybersecurity came up repeatedly during Isenberg's trip. His point was simple: every agent with access to company systems is another way in, and many companies have left those entry points unguarded.

Fleet systems hold some of the most sensitive data a company owns. Telematics shows where every vehicle is in real time, which often reveals where employees live because take-home vehicles park in their driveways overnight. Driver files include license numbers and motor vehicle records. Fuel card accounts carry spending authority. Connecting an AI tool to all of that creates a new access path that needs the same scrutiny as any other system holding employee data.

A few habits reduce the exposure:

• Grant each tool the narrowest access it needs, and separate read access from the ability to change records.
• Audit active integrations twice a year and shut off any connection tied to a tool you no longer use.
• Request each vendor's security documentation, such as a SOC 2 report, along with its breach notification terms.
• Bring IT and security into fleet technology purchases early.

Fleet technology decisions have traditionally sat with operations or procurement. As tools gain the ability to act, security teams deserve a seat at that table.

6. Separate Strong Fleet Vendors From Weak Ones

AI companies are growing revenue faster than ever, yet he heard little discussion of whether customers stick around. Fleet programs run on multiyear lease cycles and long vendor relationships, so stability matters. Before committing to an AI feature, ask how long it has been in production, how many fleets of your size use it, and what happens to your data if the feature is discontinued.

Memory belongs on the list too. One executive told Isenberg the next competitive fight will be over which AI remembers the most about its users. For fleets, the equivalent is a system that retains your policies, past decisions, and vehicle history, so its recommendations reflect how your program operates instead of a generic template.

Where to Start in 2027

None of these trends require a large technology budget to address. Most of the groundwork is operational, and fleet teams can begin this quarter:

1. Reconcile your active vehicle list across telematics, fuel cards, insurance, and lease records, using the VIN as the common key.
2. Map the weekly tasks that eat the most staff time and identify which systems each one touches.
3. Send your current vendors a short questionnaire covering integrations, data export rights, audit logs, and security documentation.
4. Draft approval thresholds that define which fleet decisions software can make alone and which require a person.
5. Invite IT and security into your next fleet technology review.

Isenberg ended his notes by observing that attention is scarce and distribution is hard for AI companies. For fleet managers, the scarce resource is time. AI tools can return some of it, but only to fleets whose data is organized and whose rules are clear.

Get Your Fleet Data Ready

EMKAY has helped companies manage their fleets since 1946, and clean, connected fleet data sits at the center of that work. Our team can help you evaluate technology partners and build a fleet program ready for what comes next. Call 630.250.7400 or email info@emkay.com to start the conversation.