The Geospatial Edge is Esri Canada’s periodic newsletter for managers and professionals tasked with growing their organizations’ geospatial capabilities. In this issue, Matt Lewin discusses five integral skills that every GIS manager needs to grow their GIS program in the age of artificial intelligence.

In 2024, I wrote about seven unique skills every GIS manager needs to succeed in their role. It was one of the most-viewed and best-reviewed articles I've ever written, and I'm glad it's had such a positive impact.

However, that piece was written during the early wave of generative AI, and at the time, it wasn't clear how AI, particularly agentic AI, would impact the role of a GIS manager.

Well, time has a way of clarifying things. And while the original seven skills I outlined remain as important as ever, the arrival of agentic AI into the GIS discipline introduces new integration challenges, budgeting considerations and team-building demands on managers that didn't exist even two years ago. Time for an update.

Here are five additional skills that every manager needs to run and grow their GIS program in the era of agentic AI.

1. Designing GIS infrastructure for agent use

Agentic AI introduces a major new design challenge for your GIS infrastructure: your system must now accommodate machines (agents), in addition to people, and these machines lack the context to navigate the peculiarities or gaps in your data or tools.

This architectural shift has practical implications across several areas:

Data readiness. Not all of your datasets are ready for consumption by an agent. Some have known quality issues, while others have informal caveats or context that only your team understands. For example, a pipeline company's integrity team, operations team and environmental team each maintain spatial layers with a field called "CONDITION," but in each case the field means something different: an inline inspection score, an operational status like "active" or "abandoned," and a surface description like "eroded" or "vegetated." A human analyst doing a dig site risk assessment wouldn't confuse these, but an agent asked to "identify pipeline segments in poor condition near the proposed excavation" lacks the context to tell them apart without clear, machine-readable documentation. You need a way to evaluate which datasets are "agent-ready" and then provide the agent with instructions for appropriate use.

Metadata and documentation. Metadata has always mattered, but there's a difference between metadata written for a GIS analyst who understands conventions and metadata written so an AI model can correctly decide whether a layer is appropriate for a given query. This is almost a new form of technical writing, involving describing your data and services in ways that steer agent behaviour rather than just inform human users.

Tool and service provisioning. When an agent connects to your GIS environment, which geoprocessing tools does it have access to? Which data layers? Which services? This is like provisioning a new employee's toolbox, except the consequences of giving the agent the wrong tools are different. A person will ask for clarification, while an agent will use whatever you give it.

Connectivity. GIS managers need to understand how agents connect to and interact with the spatial infrastructure. That includes things like API exposure and the protocols through which large language models call spatial services. You don't need to build this yourself, but you need to know enough to make informed decisions. For example, an agent that queries a published feature service is very different from one that has direct access to your geodatabase or can invoke geoprocessing tools. The scope of what it can read, write and execute depends on how it's connected. If you don't understand those distinctions, you could inadvertently introduce data access and usage risk into your environment.

2. Deciding what agents do and what people do

Five years ago, no GIS manager had to answer: which workflows stay human, which go to agents and which are a hybrid with human checkpoints along the way?* Now it's a daily conundrum!

[*Note: I wrote a manager's guide discussing this shift in detail across ten different core GIS functions. Check it out for a more in-depth read.]

Presently, LLMs and the agents they power are notoriously weak at certain spatial tasks. Precise coordinate math, topological reasoning, map projections: these are areas where agents will produce confidently wrong answers if you let them. Knowing where agents are likely to fail at spatial reasoning is a new form of technical literacy for GIS managers.

This matters because the division of labour between humans and agents is an operational design question with real consequences. If you get it wrong in one direction, you've automated workflows that need human judgment and potentially introduce a cascade of errors. Get it wrong in the other direction, and you risk inserting so many human checkpoints that you end up recreating the old manual process with a bunch of extra steps.

For each workflow, you need to assess the cost of an error. How well can an agent handle the spatial reasoning involved? Where should a human review before the output goes further?

This extends to something I think of as supervising non-human workers. You don't motivate agents or worry about their career development, but you do need to scope their authority and decide when to trust them.

For example, let's say your team handles environmental impact assessments. A colleague suggests an agent could handle the initial screening phase. But you know that screening involves subtle spatial judgments, such as determining whether a wetland boundary is accurate or based on outdated imagery.

How do you proceed? A pragmatic approach is to have the agent assemble the data package and run the analysis, and then have a human analyst review the results before it reaches a stakeholder. That checkpoint still lets you capture most of the efficiency gains from the agent while protecting against errors that could come back to bite you.

3. Managing the economics of agent-mediated work

In my original article, I discussed managing the all-in cost of GIS systems, including hardware, licensing, data, training and long-term maintenance. That skill is still essential, but agentic AI introduces a cost structure that varies from what most GIS managers are used to.

Traditional GIS costs are mostly fixed and predictable, and based on licenses and full-time staff, with a degree of cloud-based consumption costs depending on the organization's web-based setup. AI agents, on the other hand, take consumption-based costing to another level. Every query, every analysis, every output carries a cost tied to compute cycles, API calls and token-based AI services. Those costs can scale in ways that are hard to predict.

This creates three distinct economic challenges that GIS managers need to get comfortable with:

Unit economics. What does a single agent-driven analysis actually cost, and how does that compare to a human doing the same work?

Managing demand. When spatial analysis moves from a two-hour wait to a two-minute response, demand across the organization can surge considerably. Managing that (triaging, rate limiting, setting expectations) is a new operational challenge.

Building the ROI case. When an agent costs X per month, but it frees up Y hours of analyst time that you can redeploy to strategic work, what is this worth to you?

For example, say you roll out an agent that automatically generates vegetation management maps. It does in minutes what used to take an analyst two hours. It's a hit, and you expand the pilot to all field operations. Six weeks later, you realize your cloud compute charges have tripled. You failed to properly scope the volume of CPU cycles and token generation required to scale to this level. In the future, a capable GIS manager will need to understand these AI economics and establish usage guardrails before the pilot expands.

4. Assuring quality at agent scale

A particularly concerning aspect of AI I hear from managers is that agents operate at unprecedented speeds, so how do you keep up with quality assurance when an agent outpaces your team's capacity? And how do you avoid the trap of "rubber-stamping just to get it off my desk"? When a GIS team produces five maps or datasets each week, it's conceivable that a team can keep up, but when an agent generates five hundred outputs in the same period, quality assurance becomes a different discipline altogether.

GIS managers need to learn how to design QA processes that work at agent speed and volume. One method is to shift from reviewing every output to defining error rates for different use cases and then randomly sampling outputs. Different use cases will naturally have different acceptable error rates. For example, a zoning inquiry for a resident might tolerate a different error rate than an infrastructure setback calculation that informs a construction permit. The latter is more spatially sensitive and requires high precision. The process, however, remains the same: build automated checks that flag anomalies without requiring a human to review every result.

This also means building audit trails. When an agent produces a spatial output, you need to be able to reconstruct what happened. What data did it consume? What operations did it perform? What intermediate decisions did it make? What data was the model trained on? Today, when something goes wrong with a map or analysis, you walk over to the analyst's desk and ask them. In an agent-driven workflow, that conversation doesn't exist. You need observability and traceability built into the process from the start.

Quality assurance at agent scale will become ever more important as adoption of agents increases. It's what makes your agent-driven spatial work trustworthy enough to actually use.

5. Governing agents and managing risk

One of the skills I wrote about in my original article was the importance of managing and governing geospatial data. Agentic AI extends this requirement beyond data to governing agents themselves and dealing with new risks that didn't previously exist.

Key questions to answer include: who in your organization is allowed to deploy an agent that consumes your spatial data and services? What approval process do they go through? What are agents permitted to do? Are they read-only queries or can they publish outputs, modify data or trigger workflows? These are policy questions that need clear answers, and in most organizations today, nobody has written those policies yet.

Then there's accountability. When an agent produces an inaccurate spatial output, who owns that? Is it the GIS team because they published the underlying data, or the business unit that deployed it to serve the customer?

These are tough questions to answer, but as the GIS manager, you're often the best person to raise them because you understand what can go wrong with spatial data in ways that others don't. While you'll need the relevant stakeholders to weigh in on the answer, you serve as the catalyst to advance an important conversation around governance and policy related to geospatial agents.

Wrapping up

The original seven skills I wrote about remain foundational. As a manager, you still need to understand GIS capabilities, articulate value, manage costs, understand integrations, build teams, govern data and track the direction of the industry. Those skills haven't been replaced, but agentic AI adds a new dimension of complexity that demands a new focus. These skills didn't exist in any meaningful way in earlier times, but all will be essential within the next few years. If you start building them now, you won't just keep up with the changes coming to the GIS profession; you'll be leading the way.

Let’s talk 

Are you exploring the potential of AI and AI agents for your business? Are you looking to evolve your skills and understanding? Send me an email or connect with me-on LinkedIn. I’d like to hear from you! 

All the best, 

Matt 

The Geospatial Edge is a periodic newsletter about geospatial strategy and location intelligence by Esri Canada’s director of management consulting, Matt Lewin. This blog post is a copy of the issue that was sent to subscribers in April 2026. If you want to receive The Geospatial Edge right to your inbox along with related messages from Esri Canada, visit our Communication Preference Centre and select “GIS Strategy” as an area of interest.