Decoding Ourselves

1. The Longevity Project

After the computer revolution, the class turns to biology: can startups take on cancer, aging, and death itself? The framing fact is striking. For thousands of years, human life expectancy barely moved. Then, from around 1840—when it stood at roughly 45 to 46 years—it began rising at about 2.5% per decade, in a straight line that looks a lot like Moore's Law for lifespans.

• Statistically, every day you survive adds about 5 to 6 hours to your expected life.
• The U.S. actually lags the global frontier in life expectancy—there is catching up to do.
• Three futures are possible: the trend continues, stalls, or accelerates. Nothing guarantees the line keeps going; before 1840 it was flat forever.
• Biotech is the industry positioned to decide which future we get.

2. Luck, Life, and Death

One way to think about longevity is that dying is a matter of bad luck—accidents at three scales. Microscopic accidents are things like DNA mutations that cause cancer. Macroscopic accidents are car crashes. Cosmic accidents are asteroid strikes. From the 17th to 19th centuries, thinkers like Francis Bacon (in New Atlantis) imagined humanity achieving mastery over nature and overcoming accident altogether. But from around 1850, the opposite frame won: actuarial science and life insurance reduced life and death to probability functions, and indeterminacy became the dominant worldview.

• The actuarial math: a 30-year-old has about a 1-in-1,000 chance of dying in a given year; a 100-year-old, about 50%. Eventually your luck runs out.
• A telling footnote: around 1700, record-keeping was so poor that people claiming to be 150 were believed—and could earn special pensions from the King.
• The deterministic bet: if we could fix just the microscopic accidents—the mutations and cellular failures—estimated lifespans could reach 600 to 1,000 years.
• The class's core question: can biology move from the statistical, luck-driven realm into the deterministic, solvable one?

3. CS Meets Biology

Traditional drug discovery is a numbers game: screen roughly 10,000 starting compounds, watch about 5 survive to Phase 3 trials, and hope 1 wins FDA approval. Companies take 10 to 15 years to play out, with binary outcomes and little control along the way. That randomness worked for decades, but the soaring cost of each new discovery suggests the easy wins have been exhausted—which is why nearly all life-sciences funds have lost money, with returns as poor as cleantech's.

• DNA sequencing costs are collapsing: about $500 million per genome in 2000, roughly $5,000 by 2012, with $1,000 expected within a year or two.
• Yet the Human Genome Project underdelivered on its late-1990s hype—a warning that the hard part isn't reading the data but knowing what to do with it.
• Biology degrades irreversibly; computation is reversible and reprogrammable. How much of biology is really computational remains an open question—and the opportunity.

4. Three Companies Decoding Biology

The guests—Brian Slingerland of Stem CentRx, Balaji Srinivasan of Counsyl, and Brian Frezza of Emerald Therapeutics—run companies at three points on the spectrum from wet-lab biotech to pure computation. What unites them is culture: PhDs working like a hardcore tech startup, with advanced robotics, version control for lab notebooks, and heavy in-house software.

1. 1

   Stem CentRx — cure cancer

   Targets cancer stem cells, the subpopulation that drives tumor growth, instead of carpet-bombing every cell with chemo. Precise targeting allows lower, more effective doses. Mouse studies looked very promising; human trials were expected within one to two years.

2. 2

   Counsyl — screen every pregnancy

   A bioinformatics company aiming to be the default genetic test for pregnancy: one test covering about 100 genes, focused on well-defined Mendelian diseases. Already screening about 2% of all U.S. births, and built more like a software company than a biotech.

3. 3

   Emerald Therapeutics — cure viral infection

   The most computational of the three: molecular machines that tag virus-infected cells and trigger them to self-destruct, aiming at all viral infections. Operating in stealth mode, betting on a scalable platform rather than a single product.

Bio should just be sensors and gathering data. Everything else should be done at the command line.

5. Timing, Secrecy, and Talent

Thiel presses on timing: Marc Andreessen says many late-90s internet ideas were right but too early—why is biotech's moment now? Balaji compares genome sequencing to ARPANET's first packets: real, but not yet compelling; pregnancy screening is the on-ramp, and once people have their data, the marginal cost of using it approaches zero. Slingerland answers the 40-year 'War on Cancer' objection: the approach never changed—carpet-bomb chemo judged by tumor shrinkage, the wrong endpoint since tumors shrink and recur. Frezza's evidence is Genentech: founded in 1976, it opened a window in which 9 of America's 10 largest biotechs were born—and that window had been shut for 30 years before it. His bet is that a new, computational window is opening, and that clinical-trial barriers turn first movers into monopolists: imagine if Chrome had needed FDA approval to challenge Internet Explorer.

• Secrecy playbook: no press releases until a drug launches; stay quiet about techniques so rivals can't blanket-patent them (Genentech-era firms patented broad concepts and now some companies earn millions licensing patents while shipping zero drugs).
• But assume 10 hidden competitors are chasing you anyway—it forces better, faster execution.
• How to size invisible competition: sample the network like an ecologist samples a jungle; if no capital or talent in Silicon Valley is on your problem, you are probably alone.
• Recruiting under secrecy: personal referrals compound—each great engineer refers about two more, and 2^n scales. Engineers are shy referrers, so Frezza goes through their friend lists one by one asking who is good.
• Why leave web/mobile for biotech: social's flags are already planted, the genome never becomes obsolete, and it is hard to bleed and sweat for another dating app.

The next big thing won't look like the last big thing.

6. Q&A: Regulation, Feedback Loops, and Getting Started

On regulation, Thiel notes something odd: the FDA effectively bottlenecks drug development for the whole world. He sees a tipping point coming where the U.S. can no longer dictate global pace and must compete—perhaps with China—on speed, a potential paradigm shift. SpaceX is the precedent: heavily regulated at first, it persevered as aerospace rules eased. A hostile regulatory baseline can even be an advantage for those willing to endure it. On iteration speed, Frezza explains that computational biotech validates with physical models and experiments rather than external feedback, and Balaji insists slow cycles are contingent, not necessary: insulin went from discovery to patients between 1920 and 1923—software speed—and Counsyl went from conception to product in 15 to 18 months, versus a typical 7 to 8 years.

• On venture capital: 'VC is broken' with respect to biotech—funds lost money, time horizons are too short, and VCs want single-compound companies, not multi-compound platforms doing serious pre-clinical research.
• On the grind of starting: setting up a lab took Emerald about a full year of equipment-buying and troubleshooting—unlike PayPal, where the interface to physical reality was so thin that new hires just assembled their own desks.
• Balaji's closing advice: startups always begin with futons and ironing boards; what matters is getting the foundation right. You don't need a science-fair project at the start—you need to do your analytical homework.

If you want money, ask for advice. If you want advice, ask for money. That game is exhausting.

Then vs. now (2026)

Self-check quiz

Pick an answer to reveal the explanation.