A conversation with Irina Ankoudinova, biologist, genome and protein engineer, recorded August 19, 2026. Americans, she says, pronounce it Ankudinova.
Full conversation here:
The phrase she circled for three hours
I met Irina at the Sung-Hou Kim lab reunion, which is to say I barely met her. I was three conversations behind, wrangling other people’s laptops, and George Church had handed me a new deck two minutes before he went on. We talked briefly. I caught maybe a third of it, and that third was interesting enough that I went back for the whole thing.
The first part of the conversation was her career, and it had a shape. Every time I asked “and then what happened,” the answer involved a new technology arriving and making her current technology unnecessary. Zinc fingers and TALENs (Transcription Activator-Like Effector Nucleases) , then CRISPR. Sanger sequencing, then massively parallel sequencing. Structural genomics itself, which quietly lost its funding wave. She kept moving toward whatever came next, and she got hit early every time, because she worked at the edge where the hits land first.
The second part was Ukraine, where her mother’s family is from, and it had the same shape at a scale that makes the first part look small. Farmland shelled and mined out of being farmland. A language turned into a border. Families scattered across a continent and not sure when or whether they are going back.
Then, deep into a discussion about how you would even collect photographs of damaged ground, she described something that was not a data pipeline at all: a person who left and a person who stayed, each posting a picture of where they are today. Not a survey. A check-in. The landscape data would just be what fell out of it.
She never named the theme. She circled it for three hours. Displacement, then reconnection.
Part one: displaced, over and over
Third kid, outsider biologist
She was born in St. Petersburg into what she describes plainly as “a family of scientists.” Her father was a nuclear physicist. Her mother was a historian who taught high school and, importantly for everything that comes later, was Ukrainian, from Kharkiv, a graduate of Kharkiv University out of a family of teachers and doctors going back generations. Irina was the third child. Her brother and sister both became physicists.
“I’m like a, you know, outsider biologist.”
She came up through one of the strong Russian mathematical high schools, then took a degree in biology and genetics. The biology started in sixth grade. “I was always looking for answers, asking questions. Too many questions. I remember everyone telling me, it’s too many questions, ask one.” Then, with a shrug: “But I didn’t change at all.”
Because she came out of a crystallography lab, I had assumed she was trained as a biophysical chemist. She corrected me. “I feel myself more like an engineer. Genetics, there’s so much logic behind nature if you think it through the genetic way. And at the same time, I’ve always had a passion to test, to try.” Gene engineering, cell engineering, protein engineering. Structural biology, and crystallography in particular, “for me, that’s just a model.”
That line is worth sitting with. Plenty of scientists treat the instrument as the point. Irina treated it as scaffolding for a question. It is the reason none of the displacements that follow actually stopped her.
Not the doctorate track
I made a second wrong assumption on the call: that she arrived in the United States with a PhD. She corrected that one too. She had two kids in the 1990s, started two graduate programs, and finished neither. “It just wasn’t possible.”
So her path is not the academic one. It is the one where you learn a technology by running it, and the work is the credential. Worth saying out loud, given how many of us over-index on the letters after a name.
A child in her mother’s high heels
Her first paid job in the United States was in Sung-Hou Kim’s lab at Berkeley, hired as a research associate to purify protein. Then Jaru, who ran crystallization, announced she was leaving.
People outside structural biology may not appreciate what that meant. Every crystal structure starts with a crystal, and getting one is the least deterministic step in the entire pipeline. The Kim lab’s answer to that problem, sparse matrix sampling, became one of the most used crystallization methods ever published and the basis for the screens you can now buy off a shelf. I remember another Kim alum describing a Gordon Conference talk on the lab’s near-combinatorial exploration of crystallization conditions, and the sound of the room emptying afterward as people went off to make phone calls.
Those were the shoes. “Sung-Hou asked if I would learn from Jaru and stay in her slippers,” Irina said, and then, laughing at herself:
“Imagine this child in her mom’s high heels.”
She had no crystallization background. She had come out of plant biology. What she did have was molecular biology hands, cloning, vectors, PCR, sequencing, and she had already been helping run the screens. “I think he saw something in me. I really appreciate it.”
Then she described the part she loved, and her voice changed. The imager. The stacks of plates. The robot arm lifting a tray and sliding it in.
“I was in my 30s. Everything was amazing, because I’m allowed to practically do science with my hands, with such high technology. It was a good fit for my ability and my passion at the same time.”
When technology kills your technology
A little over two years in, the money went away. She remembers a large layoff in which mostly the computational people stayed. Then seven months at UCSF that she describes without decoration: “I was unlucky there.” Then she left academia for industry.
At Sangamo Therapeutics she worked on zinc finger proteins and TALENs, which were, at that moment, the state of the art for editing a genome. “It was huge. But it was very slow. We were building archives.” That work is in the permanent record, including the 2019 Nature Medicine paper on allele-selective repression of mutant huntingtin, the Huntington’s disease program, which carries her name in the author list.
Then CRISPR-Cas9 arrived. “And it was no competition with this.”
The same thing had already happened to her in sequencing. She started in the Sanger era, concatenating sequences and then demultiplexing them, and she wrote that code herself at the time (”I don’t do it anymore”). Then came Solexa, which almost nobody remembers by name because Illumina bought it, and then everything after that.
By 2017 she had been laid off again, looked around, and found DNA-encoded libraries, which she explains better than most review articles do: “When you consider DNA as just a text, you can synthesize any type of tag, link it to something, have a library of small molecules or peptides, drop it in solution, and select.” In 2017 she interviewed at Impossible Foods, the Palo Alto company a Stanford professor founded, which was screening a DNA-encoded library of small molecules to build a burger. I said their best-tasting formulation was almost certainly an accident nobody designed on purpose. She agreed instantly. Serendipity.
“It’s always disappointing when you’re laid off. You have to start over.”
I don’t think that was bad luck. I think it is the pattern. Technology cannibalizes the previous generation of technology, and the people standing closest to the frontier feel it first. It is about to happen at scale to software engineers and to most knowledge work, mine included.
Which is why the lesson she has extracted travels well beyond biology: don’t get attached to the technology, get attached to the questions you are trying to answer. Hold that. It is the only reason the second half of this piece is tractable.
Part two: displaced again, differently
The language that became a border
Then we got to Ukraine, which she handles carefully, the way people do when the subject is family.
Before the war, living here, “I didn’t have feelings that we are different. That’s why this is so devastating.” Now some Ukrainians will not talk with her, because she does not speak Ukrainian. It is not anger at her, exactly. The language itself has become the line.
“It’s mostly my generation that still speaks Russian. The young generation, this is their identity now. They don’t want to talk Russian, because they’re so offended. It’s a lot of pain.”
A pause. “It creates problems for us who want to help.”
For the record, the numbers neither of us had
We tried to size Ukraine live on the call and both failed, and she declined to guess: “I don’t want to tell any numbers.” Fair. So, checked afterward: Ukraine is the largest country entirely inside Europe, roughly forty percent bigger than California and a little smaller than Texas. Prewar population was around 41 million. After years of war and displacement, current estimates land in the low-to-mid 30 millions and are genuinely uncertain, which is itself part of the problem.
What she wanted to convey instead was that life continues. Universities are still running. She mentioned wanting to connect eventually with the Ukrainian Catholic University in Lviv. Students study, people travel back and forth, and the prewar habit of studying in Europe, the United States and Canada left networks the war has not destroyed.
The inversion nobody mentions
She also pointed out something I had not registered. Wars usually scoop up the young first. Ukraine decided not to. The fighting is being carried disproportionately by people in their forties.
“This is amazing policy Ukraine has, right? So they’re really saving their young people.”
And rather than mass infantry, drones. A swarm of five ten-thousand-dollar aircraft doing work that used to require a jet costing hundreds of millions. War is terrible and produces almost nothing worth keeping. That is one of the few exceptions, and it will outlive this war.
Soil as the mission
What Irina wants next is not a job, or not only a job. Call it a vocation. She wants to work on Ukraine’s soil.
Some of the most fertile agricultural land on earth has been shelled, burned, mined and driven over for years. Her honest position: “It’s changed every day. I cannot actually do something under current conditions.” If it happens, it happens through people who live there. Students. Universities. “That’s why in my project I want to be prepared when it’s done.”
She is not starting from nothing conceptually. There is precedent, countries wrecked in the Second World War that brought their land back, Israel’s reclamation and water recycling programs that put orchards where there was desert. “It’s doable. It’s just hard for me to find people who want to do it with me.”
And she is drawn to the biological route rather than the industrial one, which is the same instinct that runs through her whole career. Not massive capital and heavy equipment, but microbiomes, plant species matched to a specific contamination, one season of the right crop to bring a plot back to a condition where something else can grow. Maybe not wheat. Maybe beans. Then the plot becomes farmland again, or forest.
She was also honest that soil is a proxy. It could as easily be buildings. What she is actually after is the thing underneath:
“It’s motivation for the person to come and build their country, or their land, back to whatever it was. Because Ukraine is so beautiful.”
Her open question, and it is the right one: do you first build the knowledge, a map of soil conditions and what people have already done about them, or do you first find the person who will run the pilot?
Part three: what reconnection would actually look like
What followed was an hour of throwing ideas at a wall. I want to keep the shape of it rather than tidy it into a plan, because the reasoning is the useful part.
Start with what has already been photographed
Satellite and Google Earth imagery, dated but real, plus whatever drone footage can eventually be shared without exposing anything militarily sensitive. The first pass is not repair, it is triage: find the burn scars, the crater fields, the ground that has visibly stopped being soil. There is an entrepreneur I heard about who scoped an entire waste hauling business by paying someone overseas to drop a pin on every dumpster visible from Google Earth, then visiting the clusters. Same move.
Find a surrogate landscape close to home
Match the visual and soil signature of a damaged Ukrainian region to somewhere you can actually walk, a forest type in Tilden Park, say. Run controlled experiments where you are allowed to work, contaminating and treating samples systematically, then map what you learn back onto the place you cannot reach. It is crystallization logic applied to dirt: you cannot design the answer, so you set up many conditions and watch which one works.
Crowdsource the collection, automate the annotation
The best data set is the one nobody has to hand label. A phone photo already carries a geotag, and with enough of them you can train a model to infer location from the image itself, with no human curation anywhere in the loop. That is exactly how a model I built recently works. I had spent years computing expensive 3D pharmacophore fingerprints for small molecules, a minute or two per molecule, so I trained a model to predict the fingerprint directly from the flat 2D structure. It learned the conformational search. What took a minute now takes a fraction of a microsecond. The transferable lesson is not about chemistry: if you can generate labels by computation or by metadata instead of by hand, you can build a training set big enough to make the expensive step disappear.
Mine other people’s exhaust
This was Irina’s contribution and it is a good one. Construction and permitting records already encode soil condition, because you cannot build on a former gas station without declaring what is in the ground. Camera-equipped vehicle fleets record streetscapes continuously. Film crews shoot landscape for a living, and she still remembers the sunflower fields in Everything Is Illuminated. Some of that is biased, since Hollywood shoots Ukraine wherever Ukraine is cheapest, the same way Mars is usually Arizona, and she caught that herself a beat later. But the principle holds: one industry’s garbage is another’s gold.
Use the incentive people already have
Cyclists are already photographing themselves in the countryside, Irina among them. If the ask is only “post a photo, tell us roughly where,” and the mission is legible, you get real data at almost no cost. People poured ice water on their heads for a cause because they wanted to take part. I built a small app called Food Health on that same instinct: people show off what they are eating, and what falls out for free is a coarse geographic and cultural map of what the world puts on a plate. Often the background of the photo is worth more than the subject.
Build the two-sided thing, not the one-sided app
This is where it got interesting, and this is the idea the title of this piece came from. A photo exchange: someone displaced abroad and a family member still in Ukraine each post where they are today. A mutual check-in that happens to produce a landscape data set. Like a fax machine, you cannot sell just one, so the incentive to spread it is built into the mechanic.
And it sits exactly on the thing Irina kept circling all afternoon without ever naming it. Displacement to reconnection. People who left, who are not ready to go home, who still want to make something for home from a safe place.
Then she played the whole method back to me
Which is the moment I knew this was worth writing up. Take one small pilot experiment somewhere safe, in the United States if that is what is available. Solve the problem you actually want to solve. Use that output as the starting point for a model. Then keep adding data, guided by the questions you decide to ask. That is the recipe, and she reconstructed it herself before I had finished explaining it.
She flagged the obvious constraint herself. Martial law has been in force since 2022, and it is not obvious that photographing certain places is legal. “Better to check,” she said, and if it is not allowed, “then we have to omit this idea.” Which is why all of this is preparation, not action.
Part four: the tools, and who builds them
The demo I ran on her, live
Midway through, I did something slightly rude and very useful. I pasted our raw, unedited transcript into a chat model and asked it for three short stories.
Fifteen seconds later: “The Biologist Who Stepped Into the Crystallization Magician’s Shoes.” “Every Time Technology Killed Her Job, She Followed the Next Technology.” “Rebuilding Ukraine Could Start With a Photograph.” Then a fourth, “When Technology Kills Your Technology.” Irina’s response to that one was, “Actually, it is true, right?”
Her next question was the better one. Not “wow,” but why did it choose the high heels line as the center of the story. She was interrogating the tool instead of admiring it, which is most of the skill.
Scientist or engineer
Here is the distinction I keep coming back to. A scientist’s instinct is curiosity: what is that, how do we explain it. An engineer’s instinct is construction: here is a tool, let’s build the thing that answers the question.
Most people assume AI’s job is to be the scientist, that you type “how do we cure cancer” and wait. I don’t work that way. I don’t ask Claude or Codex or Gemini for answers. I ask them to build the tools I need in order to answer my own questions. It is a subtle distinction and it changes everything about what comes back. It is also why I think Anthropic has the edge right now: they went hard at code generation. The future of useful AI is not an oracle you interrogate. It is an engineer you direct. You still have to be the scientist. You still have to ask the good question.
Irina turned that around into her own frame immediately, drawn from watching her last startup fold:
“Now everything should be connected with ML tools. Whatever decision is made should be made based on broad knowledge, broad collaborations, not on whatever this small startup is doing. Otherwise you just lost.”
She has just finished MIT’s online course on machine learning and artificial intelligence in pharma and biotech, and she was clear about who it is built for: scientists and clinicians, not people who already write code. “For me it was complicated, because I had to go back to knowledge I hadn’t used for a while. Graphs, algorithms, I had no idea how it works.” What stuck was the applications, and one in particular that matters for the soil project: imaging finds patterns humans cannot.
The part where she stops needing me
She also wants to build the tools herself, which was the other thing I pushed on. She already uses Claude, mostly the free versions, in a browser. That is the part to change. Not the browser, not the chat app: the terminal. Claude Code, or Codex, or Gemini, running inside a sandboxed directory, talked to in plain language. She is on a PC and assumed a terminal still meant memorizing commands. It does not, not anymore. I have not hand-written code in almost a year, and I am a software engineer by training. She could do it in Russian if she wanted to, and she is confident she could. Her response:
“That makes me really excited. I just want to try.”
For what it is worth about why I care: the first thing I built this way was AI Dad, trained on sixty thousand emails between me and my father over two decades. It helped me get through a lot of grief, mostly the anger. That became AI Steve, which I feed continuously, text, video, photos, everything, and which gets more useful every month for exactly that reason. I run a company of twelve to fifteen people, and for roughly two thousand dollars a month in AI tools I get what I would estimate is the effective output of two hundred. That is not a projection. It is how my day works. My one running complaint is that the best model for coding and for anything touching biology is the one I am not allowed to point at biology, so I do the work a level down from where I know it could be.
Which is, if you want the tidy version, the same lesson her career taught her, pointed forward instead of backward. She survived four technologies dying by staying loyal to the question rather than the instrument. The tools she needs for the soil project do not exist yet either. So she should not wait for someone to ship them. She should build them, badly at first, the way you set up a hundred crystallization conditions and see which drop throws a crystal.
The smallest first step that actually exists
We ended on something concrete and small enough to happen this month.
I already run story circles. Toast Our Friends is the original, and there is a Toast Sung-Hou Kim edition I stood up for the reunion. Cloning one for this is minutes of my time plus a twenty-five dollar a year domain, and on the call we more or less named it: Reconnect With Family. Point it at exactly one thing, short first-person stories, anonymous if people want. What your life looks like now. What your cousin’s looks like in Kyiv or Lviv. What is unchanged. Which rituals survived. Irina’s instinct is that short beats long, and she is right, and I would add that the visceral detail is what makes a story stick, the smell of the flowers or the gunpowder in the air.
She pushed back on herself here, and it is the most relatable thing she said all afternoon: “I don’t have this type of skills to make stories short. This is such a big thing. Tell in five minutes who you are.” Thirty seconds, actually. And then, immediately: she wants to learn how, because the skill transfers everywhere.
Then cut a short out of this conversation, spend fifty or a hundred dollars promoting it into Europe and Ukraine, and watch whether anything moves. If there is a signal, register a simple domain, put a blog behind it, and let the people who showed up decide what it becomes. If there is no signal, we spent a hundred dollars to learn something. The video up top is that test, already cut.
That is what a seed looks like. Irina knows better than most that you do not design a crystal. You set up a lot of conditions, watch which drop throws one, and grow that one.
The actual call to action
Nothing here is a plan. It is a scientist with a good question, a knowledge graph that does not exist yet, and no collaborators so far.
Partway through, Irina stopped and asked me what a podcast is actually for. Her own answer is the reason this piece exists:
“Maybe if we publish this podcast, we would find somebody in the US who would be interested in participating. Especially young students. I really would like to connect to them. And I have no idea how to.”
So this is the ask.
If you are a Ukrainian student in the United States or Europe, or anywhere in the diaspora.
If you are a soil scientist, an agronomist, a microbiome person, a remote sensing person.
If you are an entrepreneur who would rather build something that eventually pays for itself than something that lives on grants forever.
If you are at a Ukrainian university and you have thought about postwar land recovery.
Irina wants to hear from you. Reply to this email, or leave a comment, and I will put you in touch.
The next step is not having the answer. It is getting five people interested in the same question.
We will talk again.
Irina Ankoudinova is a biologist and protein, cell and genome engineer whose career runs from plant biology through protein purification and crystallization in Sung-Hou Kim’s lab at UC Berkeley, zinc finger and TALEN genome editing at Sangamo Therapeutics, next-generation sequencing, and DNA-encoded library screening. She recently completed MIT’s online course on machine learning and AI in pharma and biotech. She is a serious cyclist (New Zealand, Scotland, South Korea, Japan, Belgium, Austria, Spain) and is looking for her next vocation, one she hopes involves bringing Ukraine’s land back.
Steven Muskal, Ph.D. is the CEO of Eidogen-Sertanty, Inc. - a drug discovery informatics company. He has spent four decades working at the intersection of computational biology, AI, and drug discovery. He writes about AI, health, and the intersection of biology and technology at stevenmuskal.com
For a couple mixes - here are from a recent session with daughter Hannah, Jesse, and Tim - Stand By Me and New York State of Mind. As usual, going after songs none have played together, or in case of never having played before…New York State of Mind.
