Rebuilding Incumbents with AI Beats Investing in Startups
AI's impact on the economy is uneven: restaurants and golf courses are untouched, but incumbents with brands, licenses and network effects can become market leaders once they are re-engineered.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
AI's impact on the economy is uneven
Michael Lee argues AI will not change every industry evenly. In industries like restaurants and golf courses, AI will not touch them at all; in industries like writing code, startups will win — he gives the example that Conviction and Sequence could go buy an outsourcing coding firm, which sounds like a terrible idea, but you would put your money into Cognition or Anthropic. What really excites him is the third category: incumbents that own brands, scale, network effects or regulatory moats. If you can graft the most advanced technology and engineering capability onto them, you have a chance to turn them into market leaders. Sequence's entire model is built on this judgment.
— Michael LeeBig companies cannot reform themselves
Michael uses "every company has a celebrity" to explain why incumbents cannot complete the AI transition on their own. Blackstone's celebrity is the investor, which is why it can unite the world's largest capital allocators; but in a world that believes alpha comes from engineering and AI, you need the engineer to be the celebrity who commands respect. Palantir is essentially a collection of world-class engineers wrapped in a shell and sold to organizations that cannot hire that kind of talent. If you cannot hire them yourself, going to service providers like Accenture or McKinsey does not work either — their incentive is to keep you paying, and their path is incrementalism. Buying software does not work either: software sells the workflows that already exist today, while AI requires redesigning a human assembly line that does not exist yet.
— Michael LeeA bank is the perfect cold-start testing ground
When Sequence was founded it faced a cold-start problem: without money you cannot do a deal, without a deal you cannot hire engineers, and without engineers nobody gives you money. They decided to find a customer first to prove themselves. Through an introduction from a friend, they approached a family bank in Georgia, entered as a service provider starting last August, and worked on key workflows for several months. The bank later asked whether they wanted to become a permanent partner, which became Sequence's first investment, closing in March of this year, with approvals from the Federal Reserve and the OCC along the way. Michael says banks being regulated is actually a feature, not a bug: operations are clearly defined, data hygiene is excellent, and they are very well suited to agents.
— Michael LeeInsurance brokerage is the archetypal incumbent
Sequence spent a lot of time studying insurance brokerage. In this industry more than $2 trillion in premiums flows to insurers every year, brokers do not charge customers — insurers pay them — customer retention is as high as 90%, and brokers find it hard to compete on price; these traits make it extremely difficult for startups to break in. Michael says insurers' underwriting business has barely been profitable from ancient times to today, with profits coming mainly from investing, so they have an incentive to accumulate assets, which in turn makes brokers very powerful. Baldwin meets all of Sequence's criteria: a huge market, an incumbent insulated from startup risk, and a management team already pushing technology deployment.
— Michael LeeAtlas abstracts out 80% of the commonality for reuse
Atlas is the platform Sequence built inside the bank, and they expect it to generalize across all industries. That judgment comes from an observation: if you break a business down into atomic units, 80% is highly homogeneous and 20% is vertical-specific. Atlas has four layers: the first is the data ontology, using code to define how the organization and business operate so models can understand the company; the second is the agent builder, which builds high-performance agents on top of the underlying data; the third is called lattice, the orchestration mechanism that actually connects agents into workflows; and the top layer is called artifacts, the application-building layer. Michael says all the core infrastructure they built in the bank can be reused for Baldwin and any future company.
— Michael LeeTraditional PE has structural limits
Michael acknowledges that large PE firms have the incentive, resources and capability to do AI transformation, but he thinks they face several structural limits. The first is talent: a typical PE firm is designed to make the investor the celebrity, while world-class engineers want to be part of the decisions, and it is very rare for a 25-year-old engineer at a large buyout firm to have a say in investment decisions. The second is time horizon: under a fund structure, you are usually thinking about how to package and sell in three years, which is a completely different formula from the long-term commitment technology transformation requires. The third is deal cadence: Sequence does only one deal a year, has no deployment cadence, and no LP pushing them because they did not invest enough this year. Michael says he wants to hold Sequence equity for life.
— Michael LeeThe hard numbers on the bank transformation
From the March investment to the time of recording, about 6 months, Sequence built a system in the bank that can take on all consumer lending in the organization. Average consumer loan underwriting volume fell 94% versus March. Commercial loan underwriting began rolling out in recent months, and average loan processing time dropped from 30 days to 11 days. Michael says this lets the bank take on far more loan volume with the same headcount. In this year's second quarter, the bank's loan volume doubled versus the first quarter. Historically the bank had turned customers away because middle- and back-office bandwidth was limited; now, with underwriting standards completely unchanged, the bank handled all the loan volume, and the underwriting team is smaller than before the investment — because one person retired and one person moved to the front office.
— Michael LeeBetting on people matters more than betting on ideas
Michael says if he were redesigning how to do PE today, he would only look for exceptional people solving complex problems in huge markets. From his experience at Sequence he learned that ideas are cheap, execution is very hard, and truly exceptional people always find a way to get things done — because they bring in people who cover their weaknesses, or through sheer persistence find backers to do things everyone thinks are crazy. He says everything else, whether it is the best idea you have ever seen or the hottest breakthrough of the moment, is just an important signal, but this business today is ultimately about people, especially at the early stage.
— Michael LeeIn their own words · checked verbatim
I was somehow convinced that AI would have an uneven impact on the economy.
Michael Lee3:01
In a world where you believe alpha comes from engineering and artificial intelligence, you need to create a culture where the persona known is the engineer.
Michael Lee8:01
You can't sell a product on a new human assembly line that doesn't exist today because you think so.
Michael Lee10:01
We like organizations with a fairly dense structure and centralized activities. Therefore, everything you create can be depreciated over a large base.
Michael Lee17:02
Your typical private equity firm is designed to celebrate the investor.
Michael Lee26:03
We try to do one deal a year. That's all. We have no deployment cadence.
Michael Lee27:03
Ideas are cheap, and execution is very difficult. And truly exceptional people will always find a way to make something work.
Michael Lee40:05
Figures
| Baldwin acquisition price | $7.7 billion | 0:01 |
| Sequence founded | 20 months ago | 1:03 |
| Bank annual sales | over $100 million | 17:02 |
| Insurance brokerage industry annual premium volume | over $2 trillion | 18:02 |
| Insurance brokerage customer retention rate | 90% | 19:02 |
| Decline in average consumer loan underwriting volume | 94% | 32:03 |
| Average commercial loan processing time | from 30 days to 11 days | 32:03 |
| Bank loan volume quarter over quarter | Q2 doubled versus Q1 | 33:03 |
Glossary
- perpetual holding company
- A holding structure with no fund life that holds and operates subsidiaries for the long term.
- human assembly line
- Michael's metaphor for how traditional organizations treat each person as a station on a conveyor belt.
- organizational physics
- Sequence's framework for judging whether an organization suits AI transformation, favoring dense structure and concentrated activity.
- Atlas
- The four-layer AI platform Sequence built inside the bank, comprising the data ontology, agent builder, lattice and artifacts.
- lattice
- Atlas's third layer, responsible for connecting built agents into actual workflows.
- Applied Epic
- The main operating system in the insurance brokerage industry, on which Baldwin is doing a single-instance deployment.
How to listen
Investors and founders focused on AI deployment, M&A and incumbent transformation, especially those who want to understand the logic of "don't fund startups, buy the leaders."
The founder reflections from 34:45 to 36:08, which are lower in information density.