Episode 446 of The VentureFizz Podcast features Jack O’Brien, Co-Founder & CEO of Subconscious.
AI agents are rapidly transforming knowledge work, especially in sectors like software engineering, law, and finance. However, inference costs for AI agents present a massive challenge, as agentic workloads love chewing up tokens which in return ends up costing companies lots and lots of money. And context limits or rate limits are not the answer either.
Subconscious was built to solve this exact challenge. Based on research out of MIT, Subconscious has built a specialized inference platform designed specifically for long-horizon AI agents. Using dynamic context compression and efficient caching, the platform allows developers and enterprises to run open-weight models faster, longer, and up to 80% cheaper.
It reminds me of another company that was based on research out of MIT back in 1997: local tech giant Akamai, who helped speed up the internet based on distributed computing and intelligent routing.
Subconscious recently announced $5.1 million in funding led by MassVentures, with participation from Foothill Ventures, Underscore VC, E14 Fund, Oakseed Ventures, and the Agent Fund.
In this episode, we cover:
- Jack’s background story, including being the nation’s top-ranked face-off specialist in lacrosse, his career in D1 lacrosse at Princeton, and his early entrepreneurial experience running Tiger Bed Rentals as a student through its acquisition.
- His time as a software engineer at Google and what this experience taught him.
- His serial startup journey, including building OnePager and its acquisition by Stonks.
- The origin story of Subconscious and meeting his co-founder, Hongyin Luo.
- How Subconscious’s inference runtime uses dynamic context compression to slash token costs while maintaining or even improving model performance.
- Details on their newly announced $5.1M round of funding.
- The rise of open source LLM’s versus closed frontier models.
- His personal passions outside of work, including kite surfing.
Podcast Sponsor:
This podcast is brought to you by one of the strongest longtime supporters of the local startup ecosystem, Silicon Valley Bank, a division of First Citizens Bank. With more than 1,500 bankers and relationship advisors and $44B in loans as of Q4 2025 – SVB delivers expert guidance, specialized products and a team that knows the innovation economy inside and out. Learn more at SVB.com.
Transcript:
Keith Cline: Jack, thanks so much for joining us.
Jack O’Brien: Yeah, thanks for having me on.
Keith Cline: I’m excited to talk to you because I’m very involved in the Boston tech ecosystem, and every once in a while, certain people and companies start making their presence known in a major way. I consider you and Subconscious part of that group.
We’re going to talk all about Subconscious and what you guys are building, because it is a very meaningful solution to a problem affecting almost every tech company right now.
Before we dive into that, I noticed you are very active on LinkedIn. As a recruiter my background, I signed up early on—my user number was around 55,000. Back then, it was just a directory to find people. Today, founder personal branding on LinkedIn is almost more important than a company account posting updates. What you write is compelling and thoughtful—it clearly isn’t AI slop. What is your LinkedIn strategy, and what has it done for your brand and company?
Jack O’Brien: I give a lot of credit to Bryan Benedict, founder of RC and now founder of Eliza in Boston. We had a chat where he told me, “You’re posting once a week as a founder. You need to post five days a week, ideally more than once a day. What is more cringe in your mind: your company failing because you didn’t put yourself out there, or just posting every day?”
He was absolutely right. I put my whole life behind Subconscious, so I decided to put something up every day about the team, what we’re building, or the Boston ecosystem. I never use AI to write my content because people can tell.
By refining my voice, it has been meaningful. We’ve gotten customers, hired team members, built partnerships, and received event invitations just by taking 15 to 30 minutes a day to share what we’re doing. No one cares about what you’re doing in a vacuum—you have to put it out there.
Keith Cline: Is that 15 minutes part of your daily morning routine, or do you map things out ahead of time?
Jack O’Brien: In an idealized world, I queue things up on Sunday. Usually, the hot windows are around 11:30 AM or 3:00 PM. If I’m on top of my schedule, I use LinkedIn’s built-in scheduler to queue up three or four posts for the week.
Keith Cline: What kind of content performs best for you?
Jack O’Brien: People care about stories and people. When I worked at Google, a massive portion of search traffic was around celebrities, music, and movies. People care about human connection.
When I post personal stories, photos with my co-founder Hongyin, or team updates, it performs way better than posting technical benchmarks about our system—even though our core technology is what I want to highlight. You have to capture attention through human stories first.
Keith Cline: Let’s rewind the clock. Where did you grow up, and what were your foundational years like?
Jack O’Brien: I grew up in McLean, Virginia, right outside DC. As a kid, I was pretty nerdy—big into Legos and video games. In high school, I got deeply into football and lacrosse.
By the time I was 15 or 16, I became the top-ranked face-off midfielder in the country for my class. That gave me the opportunity to play Division 1 lacrosse at Princeton.
Keith Cline: How did you become the best in the country at such a specialized position?
Jack O’Brien: As a freshman in high school, I made the varsity team and wanted to get more reps. I set up a system on GarageBand where I recorded my own voice saying, “Down, set, whistle,” with randomized delays between the commands.
I created 10 to 20 variations and went out to practice 50 to 100 face-off reps every single day for four years. Nobody else was putting in that level of deliberate practice, so by my sophomore and junior years, I was dominating. That dedication taught me that you can become world-class at a specialty if you build a system and put in the reps.
Keith Cline: You also started your entrepreneurial journey while playing D1 sports at Princeton. How did that happen?
Jack O’Brien: During college, a friend on the lacrosse team and I launched Tiger Bed Rentals. 95% of Princeton students slept on uncomfortable campus-issued twin beds. We approached Sleepy’s, told them we were outfitting a “hotel,” and bought comfortable mattresses in bulk at a huge discount.
We set up a website offering students a full-year mattress rental and extender for $350. I scraped the student email directory, sent a email blast, and collected all the pre-orders before we even bought the inventory. We ran the business through our junior and senior years, stored mattresses in temporary warehouses over the summer, and eventually sold the company to an aggregator rolling up campus service businesses.
Keith Cline: After Princeton, you started your career at IBM and then landed at Google. What was that experience like?
Jack O’Brien: Working at Google was a dream job. I spent three years as a software engineer working on Google Search, specifically within the education and search verticals.
I built features for math queries—like interactive solvers for quadratic equations—and worked on integration across YouTube. It was incredible knowing code I wrote hit tens of millions of users daily.
However, at a company that size, you move at a glacial pace to protect the core business. I missed the fast-paced build cycle of startups, which drove me to work on side projects that ultimately led to my next venture.
Keith Cline: That next company was OnePager. What was the story behind building and exiting that business?
Jack O’Brien: My bed-rental co-founder Adam Hardy and I initially built a scout network called Open Scout across 50+ universities to connect student founders with VCs. To standardize how students shared company profiles, we built a simple, interactive deck platform called OnePager.
When COVID hit, remote fundraising exploded, so we opened OnePager to the public. Founders used it as a modern alternative to static PDFs, while investors paid for access to our database. We also became one of the top lead referrers for AWS Startups and Stripe. In early 2022, Stonks acquired OnePager, and I served as CTO through the integration.
Keith Cline: What made you decide to pursue your MBA at MIT Sloan?
Jack O’Brien: In mid-2022, GitHub Copilot and ChatGPT made it clear that AI was going to transform software engineering. I applied to MIT Sloan to transition from CTO to CEO and find a world-class technical co-founder to build a foundational AI infrastructure company.
While at MIT, my team won the MIT $100K Pitch Competition with Max Compute Co., a prompt-routing engine. Though prompt routing was a hot space, we realized margins were thin and competition was crowding in, so we pivoted away. Shortly after, an MIT professor introduced me to Hongyin Luo, a PhD candidate at CSAIL. We hit it off immediately and co-founded Subconscious.
Keith Cline: What is Subconscious, and what problem does it solve?
Jack O’Brien: Subconscious is an inference platform built specifically for long-horizon AI agents.
When developers build complex AI agents—like coding assistants, legal researchers, or enterprise automation tools—those agents run multi-step workflows that generate hundreds of thousands or millions of tokens across hundreds of discrete tool-calling steps. Standard LLM inference platforms treat agentic workloads the same as simple chatbots, resulting in extreme latency, memory bottlenecks, and massive API costs.
Based on Hongyin’s research at MIT, Subconscious built a specialized runtime layer that uses dynamic context compression and efficient caching. We dynamically compress an agent’s context over long trajectories without losing crucial reasoning history.
This allows developers and enterprise teams to run open-weight models faster, longer, and up to 80% cheaper, while capping the GPU memory footprint per agent. It acts like an Akamai CDN for long-context AI inference.
Keith Cline: What types of customers are adopting Subconscious today?
Jack O’Brien: We serve two primary customer segments:
- Enterprise GPU Clusters: Large engineering teams running open-weight models on their own GPU infrastructure. Deploying Subconscious acts as a 2x to 3x force multiplier on their existing GPU clusters, dramatically increasing concurrency and speed.
- Agentic Software Companies: Companies building token-heavy AI agents for software development, legal, finance, and enterprise workflows. One of our beta users—a 20-person engineering team—was spending $40,000 a month on closed model APIs. By switching to Subconscious’s managed inference platform, their bill dropped to $6,000 a month while their developers reported faster token generation speeds and higher benchmark performance.
Keith Cline: You recently announced a $5.1 million funding round. Who participated in the round?
Jack O’Brien: The round was co-led by MassVentures in Boston and Foothill Ventures in San Francisco, with participation from Underscore VC, E14 Fund, Companion Ventures, Oakseed Ventures, and the Agent Fund. Having strong investor representation in both Boston and Silicon Valley has given us an incredible network across both coasts.
Keith Cline: What are your thoughts on the debate between open-weight models versus closed frontier models?
Jack O’Brien: Open-weight models are going to dominate enterprise adoption. Closed models from OpenAI or Anthropic are great for testing, but enterprises want to protect their proprietary IP, customize models through post-training, and avoid exorbitant API bills.
Today, open-weight models like Qwen, Llama, and GLM are performing near top-tier frontier levels at a fraction of the cost. As agentic workloads scale to trillions of persistent tasks, open-source infrastructure powered by platforms like Subconscious will be the foundation.
Keith Cline: What do you enjoy doing for fun outside of building Subconscious?
Jack O’Brien: While at MIT Sloan, I learned to kite surf! Spending weekends along the Massachusetts coast—at the Cape or Nahant—has been a great way to disconnect. Learning it involves getting dragged through the water for three days, but once you figure out how to lean back and let the wind take you, it’s an incredible rush.
Keith Cline: Jack, thanks so much for walking us through your story and sharing what you’re building at Subconscious.
Jack O’Brien: Thank you so much, Keith! Really appreciate it.
