One of Boston's most influential techies is working on a next big thing: AI ‘agents'

Boston Globe

31-01-2025

If you've tried using the Copilot AI feature inside of a Microsoft application, 'that is an AI helping the human,' says Mark Roberge, a former HubSpot executive who now teaches at Harvard Business School. (Copilot can do things like help you summarize or condense lengthy content, or generate images for a slide presentation.) 'Agentic AI moves the AI to the front, and the humans are architecting it. ... It's a different paradigm, and it gets closer to something that's truly disruptive,' he says.
Advertisement
While Shah and Cambridge-based HubSpot only began promoting the Agent.ai site last year, it has already attracted 230,000 users who've set up accounts. And 250,000 people have signed up for an email newsletter Shah oversees.
The site is intended to serve as an agora for agents. Right now, everything on it is free, but Shah says there's the possibility of adding payments in the future. You can find agents that will design logos, write cover letters based on a job description, create transcripts of YouTube videos, or draft subject lines that are likely to increase the chance someone opens an email. There are more than 280 different agents listed on the site; Shah personally created 17 of them. Each one connects existing AI services on the internet like Lego bricks to make something new.
Advertisement
Shah says the most popular, so far, is one he created called Company Research Agent. You enter the name or web address of a company, and it compiles a dossier for you. Shah says that when you query it, it gathers info from a half-dozen different services, including Google, OpenAI, HubSpot, and Perplexity, an AI-powered search engine.
The Company Research Agent on agent.ai takes the name or web address of a company, and it compiles a dossier for you.
SCREENSHOT OF AGENT.AI
'That one was a case of scratching my own itch,' Shah says. When he'd assess potential partnerships for HubSpot, or startups that he might invest in, 'I used to do it manually, looking at website traffic, who are the founders, what are their backgrounds.' With the AI agent, he says, he has reduced a few hours of research to about 60 seconds.
The reports produced by Company Research Agent are impressive at first glance, but like anything spit out by today's AI services, you need to do a fair amount of fact-checking. When I asked for a report on Agent.ai, it confused Shah's site with an earlier startup company that once owned that web domain. It listed a Silicon Valley address, a founding year of 2016 (Shah started working on his version of Agent.ai in 2023), and described a product that used AI to provide customer support. Same with the Boston Globe: It described a newspaper still owned by the New York Times Company — which was accurate from 1993 to 2013 — and which has either 5,000 or 1,200 employees. (The correct number is 'nearly 1,000,' according to Globe spokesperson Carla Kath.)
Advertisement
Despite some accuracy challenges, it's very good — not to mention fast and free — at serving up information about a company's product, pricing, executives, media coverage, web traffic, and funding. I had decent luck using other agents to create logos and visual memes about next month's Super Bowl. But some of the agents, like the cover letter generator, just didn't work after multiple attempts.
Paul Baier, cofounder of the Boston analyst firm GAI Insights, says he has been a user of the Agent.ai site, which he calls 'a very interesting experiment.' Agents, he says, will let people get access to a set of 'best-in-class' tools to do a particular job — he uses grammar and spell checking as an example — rather than just the tools built by one software company into their own application.
'Digital work will look radically different two years from now,' Baier says.
If that prediction plays out, it could affect every software company — including HubSpot, which sells software to help companies communicate with and market to customers and prospective customers. The consulting firm CapGemini issued a report last summer that said 82 percent of companies expect to be using AI agents within three years.
Agent.ai was initially created as a project of an LLC that Shah owns. He worked with freelancers to build it — not HubSpot employees — because Shah says that afforded him the ability to move fast, and not necessarily need to tie in to other HubSpot projects or product roadmaps. As an example, he quickly integrated the newest open source AI model from
Advertisement
A challenge with companies of HubSpot's size, he says, is: 'How do you experiment? How do you try things without distracting the core team?' (Shah started a similar side project to explore the potential of AI chatbots, called ChatSpot. HubSpot acquired that project for $1, and Shah says it could eventually do the same with Agent.ai. He says he's not building the site for 'personal profit or gain.') But as of late January, Shah said it was possible that the company would take a more active role with Agent.ai, and 'put a full HubSpot team on it later this year.'
On agent.ai, you can find agents that will design logos, write cover letters based on a job description, create transcripts of YouTube videos, or draft subject lines that are likely to increase the chance someone opens an email.
SCREENSHOT OF AGENT.AI
Roberge, a former top sales executive at HubSpot, calls Shah 'an awesome visionary. He's had free range to do these things that explore new technologies. And he's the most loyal HubSpotter in the history of the world.'
Shah and Brian Halligan started the company in 2006. While there was
Baier, the analyst at GAI Insights, calls AI agents 'a really hot space' and says he hasn't seen other sites like Shah's that are trying to collect or curate them. In January, ChatGPT introduced an agent of its own called Operator, in a 'preview' mode.
Advertisement
For his part, Shah says his goal with Agent.ai is simple: for 'the world to be able to use agents, and understand them better.'
Scott Kirsner can be reached at