iia/1130 Built in Boston Scaling Global AI Company

• Built in Boston: Scaling a Global AI Company

  • meta

    • How Jellyfish scaled globally by embedding generative AI into creative, media, and data — and what that journey says about building global tech from the MA ecosystem.
    • [Panel]
    • [Building Intelligence]
    • 11:30 AM
    • moderator Person/Sarah Krouse
      • wsj The Wall Street Journal
        • Technology & Media Editor
    • panelists
      • Person/Brian Elliott
        • Blitzy
        • Co-Founder & CEO
      • Person/Andrew Lau
        • Jellyfish
        • Co-Founder & CEO
  • discussion
    • Person/Andrew Lau
      • last year less than 50% were doing AI Coding
      • now it's greater than 80%
    • Person/Brian Elliott
      • moving from Engineer/ing being the Bottle/neck to Product teams being bottleneck
    • what about at the human review level?
      • Person/Brian Elliott we need Taste
    • can humans keep up?
      • Person/Brian Elliott test coverage, TDD
      • Person/Andrew Lau implicit here is Role transformation
        • the impediment is that we are going to change jobs
        • there used to be split between product and engineers, now there's no code writing, so maybe half our jobs are obsolete. Architects vs Product Managers, who wins?
        • here are the new Roles - we are unbundling and rebundling of roles
          • a role is a collection of Tasks and Jobs.
          • our titles and definitions have to change, and that's not a thing for organizations to go through
        • the act of Coding is no longer there, so the coding Roles are going away
          • the line engineer is going away. Everyone is suddenly Arch/itect
    • as adoption speeds up, are we heading towards more Layoffs.
      • Person/Andrew Lau
        • everybody is making so many things, can your customers really absorb those things? Question/Good
      • Person/Brian Elliott
        • we are ultimately constrained by creativity not a bayesian transformer architecture
    • how do your customers measure ROI?
      • Person/Andrew Lau
        • let's look at how much value is created
        • look across the entire system, not just amount of code created
        • Person/Brian Elliott says that people at Jellyfish might be the best to talk about this
    • we've heard about Token/Max/ing
      • Person/Andrew Lau
        • the times just had an article about token maxing
        • think of a Diagram/2x2
          • who is spending tokens and not doing anything?
      • Person/Brian Elliott
        • the "cold call" maxers comparison - just because you are calling a lot of people doesn't mean you're the best salesperson
        • token maxing is a metric, but it's messy
      • Person/Andrew Lau
        • Salary vs Tokens
          • companies are trying to allocate
            • which teams can use which models
            • there are Trade/Offs
              • companies are allocating $1000 / month for engineers for AI Coding
        • desire of want to use these things, but "upside down by accident" Confused
    • where are these resources coming from?
      • Person/Andrew Lau
        • we all watched 2025/10 to january 2026/01 and they are all reacting
        • here's a riff
          • we need to separate marketing vs r&d vs etc
          • these are significant
      • Person/Andrew Lau
        • Person/Jensen Huang said, Quote:

          • if a $500k engineer isn't spending $250k in tokens, something is wrong
            

        • 10 years ago, CTO, CFO, COO would bicker once per year about headcount allocations to meet CEO vision
          • today, it's every 10 days, because it's about Token allocations
      • Person/Brian Elliott
        • we are evaluating AI/Model/Providers every day full time
        • OpenAI is amazing at Structured Output
        • Anthropic is good at entropy
          • TODO what does he mean by that? My Question
        • Gemini is better at effective context window
          • there's advertised context window, but google is good at actually expanding the context window
          • I haven't heard this before; Tip/Great
      • Person/Andrew Lau
        • these tools are very volatile
        • 9 months ago it was CursorAI and now it's Claude/Code
        • at the very least, the way people are viewing Claude/Code as a command line thing they can dispatch
        • in the IDE cursor is more dominant
        • all these tools allow you to do picking of AI/Models
        • CursorAI is starting to use Open Source models in their training
        • Codex/CLI is also going there (they support it)
        • people figure out what works in Word/of/Mouth
        • every time we think we have a pattern, then boom, there's another idiom that shows up.
        • think about autonomous tools that go and submit PRs.
        • AI/Agent/Agentic/Engineer/ing/Autonomous autonomous agents - top companies 15% are autonomous PRs, last year it was 0.x%! transfor
        • until we see that second order derivative change all cards are on the table
      • Person/Brian Elliott
        • there are mid-market pure bespoke tools
        • there are also very solely for the Enterprise tools
      • Person/Andrew Lau
        • labor force is a big factor here.
    • what's the next pocket of transformation
      • Person/Andrew Lau
        • Anthropic/App/CoWork