Data-Smart City Pod – Episode Summary
Episode Overview
Title: Agentic AI Comes to City Hall
Date: March 11, 2026
Host: Stephen Goldsmith (Bloomberg Center for Cities, Harvard University)
Guests:
- Santi Garces (CIO, City of Boston)
- Mitch Weiss (Professor, Harvard Business School; former Boston Chief of Staff)
This episode explores the arrival and impact of agentic AI in municipal government, focusing on tangible experiments and philosophical shifts needed to unlock AI’s full potential in cities. The discussion ranges from successful (and less successful) pilot projects in Boston, to broader questions about adoption, organizational capacity, technical architecture, and the interplay between the public and private sectors.
Key Discussion Points & Insights
1. Principles for AI Adoption in Cities
-
Experimentation is Essential
- Mitch Weiss emphasizes the “build, test, learn” approach from 20 years of government innovation:
“Now that we're in the world of AI, which is built on the world of machine learning, experimentation is absolutely key.” (01:45 – Mitch Weiss)
- Continuous improvement and iteration are central to successful AI integration.
- Mitch Weiss emphasizes the “build, test, learn” approach from 20 years of government innovation:
-
Thinking at Scale, Not Just Pilots
- Weiss warns against limiting goals to minor service improvements:
“The AI moment in cities is one where...we don’t go, how do we, like, help 10 more people get their permits done a little bit faster? But we go, like, how do we help everybody get their permits done instantly?” (02:45 – Mitch Weiss)
- Weiss warns against limiting goals to minor service improvements:
2. Boston’s AI Experimentation: Successes and Lessons
-
AI Semantic Search at Boston.gov
- Santi Garces recounts revamping the city website’s search with an AI-powered semantic search.
- Result:
- Increased user satisfaction from 10% (old system) to 34% (AI system).
“There’s not a lot of things that make government work 3.4 times better. Let’s just switch it.” (05:18 – Santi Garces)
- Lesson Learned:
- The new experience benefited the public, but staff found it less intuitive, revealing the need to balance constituent and internal user needs, potentially with separate interfaces.
-
Scaling Beyond Pilots
- Importance of collecting user data even during pilots to inform eventual organization-wide rollout.
- Recognizes the danger of optimizing city systems for internal workflows at the expense of residents’ experiences.
3. Barriers and Strategies for AI Adoption
-
The "Growing Gap": Tech Capacity vs. Organizational Absorption
- Mitch Weiss notes rapid AI capabilities outpacing organizations’ ability to integrate them:
“Why do we have this technology capacity but only this organizational absorption is like the riddle of our era.” (08:33 – Mitch Weiss)
- Mitch Weiss notes rapid AI capabilities outpacing organizations’ ability to integrate them:
-
Addressing Concerns—Openness, Safety, and Experimentation
- The need to honestly acknowledge risk and ethical questions, not oversell AI as risk-free.
- Getting AI tools directly into employees’ hands, rather than over-focusing on training, diminishes doubts and fosters adoption.
“We’ve over indexed on training...and we’ve under indexed on putting the tools in their hands to invite them to good use.” (09:43 – Mitch Weiss)
4. The Concept and Promise of Agentic AI
-
Defining Agentic AI
- Not about a strict binary (“agent or not agent”), but increasing “agentic” capacity—more memory, execution, sensing, and collaboration capabilities.
“Let’s not imagine we have to create some archetypal agent, but just say take what we're doing and make it more agentic.” (10:59 – Mitch Weiss)
- Not about a strict binary (“agent or not agent”), but increasing “agentic” capacity—more memory, execution, sensing, and collaboration capabilities.
-
MCP Servers: Unlocking Consistency and Security
- Santi Garces explains Model Context Protocol (MCP) servers as intermediaries between AI systems and city data/APIs, ensuring consistent, reliable, and secure interactions.
- Use case: Using MCP to let AI find and synthesize open city data for more actionable insights.
“An MCP server allows...the behavior of the tool to be more reliable and to follow a particular pattern…more consistency, more secure.” (12:09 – Santi Garces)
- Collaborative Knowledge Sharing:
- Skills developed by individuals (like Santi) can be packaged and reused in other agencies, helping scale innovation much more easily than before.
“We actually have these ways now of making Santi available or Santi's...knowledge...available to all the other CIOs in the world.” (16:36 – Mitch Weiss)
- Skills developed by individuals (like Santi) can be packaged and reused in other agencies, helping scale innovation much more easily than before.
5. Human+AI Roles and the "Personas" Concept
-
From Agents to Co-Workers
- Discussion of whether anthropomorphizing AI as “co-workers” is a useful metaphor.
“Some of the AI companies are trying to get us to do this…OpenAI calls their agents at work...co workers.” (19:09 – Mitch Weiss)
- Value in developing AI "question-askers" or agents focused on surfacing risks, anomalies, or alternative stakeholder perspectives.
- Discussion of whether anthropomorphizing AI as “co-workers” is a useful metaphor.
-
Agentic AI for Stakeholder Representation
- The future may involve dedicated agents representing overlooked constituencies at meetings, enhancing inclusion and foresight.
6. AI’s Impact on Software Development and Procurement
-
Democratization of Coding
- AI tools double productivity for developers and democratize coding (potentially allowing any city employee to build apps).
- Raises questions about maintaining standards, security, and manageability amid increased “citizen development.”
“What happens if everybody in city hall could become a coder?...Our role becomes how do we enable for people to be able to deploy these things?” (24:48 – Santi Garces)
-
Resiliency as a Hidden Superpower
- AI’s rapid-build capabilities mean cities can develop their own tools in emergencies or when products fail, boosting resilience.
“Think about what it means now that...anybody in any city, if they need a piece of technology...they can spin it up quickly for resiliency.” (26:20 – Mitch Weiss)
- AI’s rapid-build capabilities mean cities can develop their own tools in emergencies or when products fail, boosting resilience.
-
AI Procurement Risks
- Even with rapid prototyping, data security and vendor lock-in remain major considerations.
“The biggest risk if you're going with some new vendor you don't know enough about is what's happening with your data.” (27:32 – Mitch Weiss)
- Flexibility and competitive alternatives are necessary for robust procurement strategies.
- Even with rapid prototyping, data security and vendor lock-in remain major considerations.
7. Advice for Cities and Sector Comparisons
-
Advice for Other Cities
- Try more; don’t get paralyzed by fear of the unknown or stuck at the pilot stage.
“There's a type 2 error in government. There are things that we should be doing that we're too afraid to try. We should be trying to do things.” (29:54 – Santi Garces)
- When something is proven, push it beyond pilot to scale.
- Try more; don’t get paralyzed by fear of the unknown or stuck at the pilot stage.
-
Public vs. Private Sector Lessons
- Leadership from the top is essential, as is measurement and having strong data foundations.
“Those that were data ready in the first place have been able to take advantage much swifter...” (32:09 – Mitch Weiss)
- Government must also keep pace, as “nation states are going to be pitted against each other...for their populace, systems to be supported is going to be more important than ever.” (33:29 – Mitch Weiss)
- Leadership from the top is essential, as is measurement and having strong data foundations.
Memorable Quotes & Moments
-
On Agentic AI Philosophy:
“Let’s actually take the AI stuff we've been doing and say can we make it more agentic?”
(10:57 – Mitch Weiss) -
On Unexpected Impacts of AI Pilots:
“We’ve made a search experience that is better for constituents. But...we started to hear a lot from our colleagues in City Hall...the search terms that I was used to searching are not working.”
(05:38 – Santi Garces) -
On Democratizing Development:
“What happens if everybody in city hall could become a coder?...Our role becomes how do we enable for people to be able to deploy these things?”
(24:48 – Santi Garces) -
On Leadership and Urgency:
“You do actually need people from the top who beat the drum that, like, we have got to change.”
(32:13 – Mitch Weiss) -
On Using AI for Unintended Consequences:
“You could just have your unintended consequences agent, like a thing on your shoulder...at every meeting go here, you know, have you thought about the consequences?”
(21:09 – Mitch Weiss)
Key Timestamps
- [01:45] — Weiss on experimentation and scale as principles
- [03:23] — Garces details Boston’s AI search pilot and scaling lessons
- [07:42] — Weiss on technology’s exponential growth vs. slow organizational adoption
- [11:47] — Garces explains MCP server (Model Context Protocol) and implications for city data
- [16:26] — Weiss on sharing AI “skills” across cities
- [18:45] — Weiss on AI “personas” and question asking agents
- [23:11] — Garces on increased productivity and future role of CIOs
- [27:25] — Weiss and Garces discuss procurement, competition, and data security
- [29:41] — Garces’s advice: embrace experimentation; move good pilots to scale
- [31:45] — Weiss: lessons from private sector—leadership, measurement, data readiness
- [33:23] — Weiss: on global competition and the need for government agility
Takeaway
This episode offers a rich, candid exploration of how agentic AI is transforming city governance, both in pragmatic detail and philosophical orientation. The conversation moves from tangible advances in Boston to broader lessons on risk, leadership, experimentation, and the future architecture of public services—emphasizing that the true power of AI lies in scaling benefits to all residents, not just incremental departmental wins.
