Summary

Process quality—not a single year’s return—is what can actually improve an investor. Walker finds motivation in seeing someone conduct better research; Fokin still values comparable long records, arguing that roughly 25% CAGR for 20 years is probably evidence of exceptional skill. Both resist “scoreboard, bro” conclusions from concentrated records where a handful of bets may explain everything.
Compounder investing means deliberately betting against broad base rates, then demanding enough upside to make the asymmetry work. Most companies will not compound revenue or earnings at 20% for 10 years, so investors must improve the odds through filters such as owner-operators and attractive unit economics. Fokin’s second requirement is venture-style convexity: the winners must have enough upside to offset the many bets that will not work.
Fokin warns that AI might make markets more consensus-driven by encouraging investors to outsource the intellectual work that creates conviction. Walker argues that similar questions posed to similar models may produce similar answers, while Fokin says the ability to build a mosaic from 15 transcripts or numerous expert calls could atrophy like navigation after GPS. The winners will use AI “as a supplement” and recognize when its polished output creates “fake knowledge.”
As AI and expert libraries democratize information processing, edge may migrate toward creating information that was never online. A large fund once had an enormous advantage because it could finance extensive bespoke research; now smaller firms can translate filings, search libraries, and synthesize calls cheaply. The remaining opportunity is to uncover neglected companies, conduct customer work, attend industry events, and realize that “everybody else is running left” when the evidence says to run right.
Fokin’s biggest process change has been a much heavier focus on customers and the difference between satisfaction and genuine advocacy. His painful 2015 Agrify investment might have been avoided by discovering that customers valued the product but were not “raving fans.” Yet Walker’s pushback survives: customers can dismiss an unlaunched product and embrace it two weeks later, so calls are inputs—not rules.
A respected friend’s idea deserves priority in the research queue, not a lower underwriting standard. Fokin’s objective is to “make the conviction my own” by completing the same work—transcripts, customer calls, management questions, and other routines—regardless of who sourced the name. Investors should also weight tips by the source’s actual superpower: an obscure $200 million company may fit Fokin’s pattern recognition; a coal special-dividend trade probably does not.
Expected IRR cannot force-rank a portfolio unless it is explicitly penalized for leverage and existential risk. Walker finds raw IRR estimates pull him toward the most levered equities, while enterprise-value comparisons pull him toward cash-heavy companies with little upside torque. Fokin’s answer is judgment: two stocks offering 25% expected IRRs over three years are “not born equal” if one is debt-free and the other has 6× EBITDA, 4× debt, and 2× equity in the example.
Expert-call libraries and AI are a “match made in heaven,” provided the model accelerates learning rather than deciding the investment. Walker’s workflow moves from an LLM synthesis of five calls, to individual call summaries, and finally to full transcripts, improving both comprehension and retention. Fokin maps AI onto “digging, analyzing, deciding” and asks where it belongs heavily, sparingly, or not at all.

Deep dive

1. Investing improves through craft benchmarks, not annual scoreboards

Fokin names the episode’s governing idea “perfecting the craft”: investing is a profession improved through iteration, feedback loops, conversations with other practitioners, and occasionally “cloning or stealing their ideas”—meaning process ideas, emphatically not stock tips.
Walker contrasts investing with racing, where a faster finishing time plainly establishes who performed better. Seeing someone break a physical barrier changes what others believe possible; in investing, a 300% year versus a 4% year reveals almost nothing without knowing concentration, options exposure, leverage, and luck.
What motivates Walker is hearing that another investor conducted five expert calls, uncovered a neglected fact, and followed it through a productive research rabbit hole. That identifiable piece of craft makes him think, “I wish I had that insight,” and gives him something actionable to improve.
Fokin finds long-duration records more inspirational when the style is comparable: approximately 25% CAGR for 20 years probably belongs among the profession’s best. Renaissance Technologies’ Medallion Fund is intellectually fascinating but plays a different game; Joel Greenblatt’s published record is a more relevant benchmark.

2. Strong returns can signal skill without ever proving it conclusively

Walker’s challenge is statistical significance: a concentrated manager with a strong 14-year record but perhaps only 10 meaningful positions could still resemble the lucky coin-flipper. Buying the Mag 7 early—or concentrating in Tesla—could manufacture legendary-looking results from remarkably few decisions.
Fokin’s honest non-answer is that the question may be unanswerable at 95% confidence and partly becomes “an issue of faith, of belief.” He suspects investors often call people brilliant when their reasoning resembles their own, while dismissing unfamiliar styles they do not understand.
Investment talent also has multiple forms. One person may be a superior business analyst, another exceptional at idea generation, another at risk control; Fokin’s least explicable category is the investor with a persistent “money smell” who seems to understand a position less deeply yet repeatedly makes money.
The “superpower” framing is relative, not absolute: it means the strongest component of someone’s own toolkit. A useful network therefore contains people with different superpowers, especially those whose strengths feel foreign enough that one might otherwise misclassify their success as luck.

3. Bitcoin separates asset allocation from conventional security analysis

Walker stress-tests track-record worship with Bitcoin’s move from roughly $200 to $115,000, more than 500x in his framing. A holder from 10 years earlier may have outperformed every conventional investor they know, but that return alone cannot reveal whether the original decision was intelligent.
Fokin distinguishes investing from asset allocation. With a hypothetical $3 million net worth, committing $2.8 million would resemble Russian roulette: “It also statistically works, but it’s a dumb idea to play it.” Committing $300,000, or 10%, could indicate excellent allocation and an ability to envision alternative future states.
Even then, it remains a “statistical sample of one.” The profitable future state played out spectacularly, but Fokin preserves the distinction between responsibly buying exposure to a possibility and risking nearly everything on it.
Walker’s frustration is psychological: he wants definitive endings, while investing rarely supplies them. Unless one is Renaissance Technologies running 1,000 trades per day, the craft demands decisions under ambiguity rather than proof that a particular process was unquestionably correct.

4. Compounder investors must combine hostile base rates with explosive upside

Fokin defines the relevant compounder as a company with attractive unit economics, a long growth runway, strong management, and demonstrated execution. Yet most businesses will not deliver 10 years of 20% revenue or earnings growth, so this style begins by betting on outliers.
Investors can recut the universe using criteria such as a founder-owner with at least a 20% stake, specific margins, growth, or returns on capital, and backing from respected venture firms. These filters might improve the odds, but Fokin doubts they make the base rates overwhelmingly favorable.
Walker invokes research suggesting a tiny minority of stocks creates most long-run market wealth, while noting possible survivorship limitations. Fokin’s defense is the intellectual 80/20 rule: an imperfect study can still preserve the essential insight that a small group of winners drives aggregate returns.
Because success is statistically unusual, Fokin wants “convexity” on the upside—even if that term is not mathematically exact here. Drawing on Zero to One and The Power Law, he argues that public-market investors can learn from venture capital’s focus on power-law outcomes without becoming venture capitalists.

5. AI could make markets more consensus-driven, not necessarily more efficient

As of early August 2025, Fokin’s deliberately provisional concern is that AI might make markets “dumber,” by which the discussion ultimately means more consensus-driven. He is not a futurist and could change his mind tomorrow, in three months, six months, or 12 months; Walker separately says he is not a coder or software developer.
Fokin’s GPS analogy carries the concern: older generations may navigate cities better because the mental muscle was repeatedly exercised. If younger investors outsource research, reasoning, and conviction formation to AI, the corresponding investing muscles may similarly “go away.”
Walker argues that similar analysts asking similar questions of the same leading models will receive broadly similar answers, notwithstanding differences in prompts and temporary model quality. Opportunity then grows where models miss a critical datum or the relevant information was never written down.
Walker reframes the history of alpha: calculating low P/E ratios once mattered, computers arbitraged that away, and understanding Google-like unit economics and runways became more valuable. Fokin accepts this as another angle but refuses to predict whether AI will specifically eliminate short-, medium-, or long-term alpha.

6. AI compresses information advantages while creating the danger of fake knowledge

Fokin draws a sharp distinction between reading a polished 30-, 40-, or 50-page AI output and personally working through 15 earnings or conference transcripts, live presentations, and expert calls. The first may impart facts without the deep, slowly assembled conviction of the second—what he calls “fake knowledge.”
AI also erodes existing competitive strengths. An outstanding writer remains better than average, but average writers assisted by AI have closed much of the gap; investors must therefore identify which personal advantages are becoming obsolete and reinvent themselves around new ones.
The broader trend is democratization. A $2 billion fund could once spend roughly $1,000 per bespoke expert call while a $5 million fund could not participate; searchable expert libraries and cheaper calls may compress a hypothetical 99- or 100-unit resource gap to perhaps 20 or 30.

7. The next edge may come from creating information the internet lacks

Walker’s candidate for the next advantage is field-generated evidence: non-obvious datasets, relationships built with 15 participants at a lumber conference, or other lawful information that no model can retrieve because nobody has put it online. Fokin says they are “looking in the same direction.”
Fokin’s ideal small-cap search result is not dozens of expert calls but zero—or perhaps one old call—plus an invitation to commission one. Sparse written information makes AI less independently useful because “language is the key word”: a large language model needs source language from which to reason.
His example is Sofwave, an Israeli-listed maker of skin-tightening devices with recurring razor-and-razor-blade economics; he disclosed that Kakuna Capital LLC and its affiliates own shares. Its Hebrew financials could be translated by AI, but the initial expert library contained only one somewhat negative former-employee call.
Fokin believed Sofwave had already addressed that employee’s concerns, then supplemented the record through doctors, who buy the product, patients anecdotally, and former employees covering culture and sales. Once eight or nine calls accumulate, AI can accelerate synthesis—but early discovery and proprietary qualitative work created the information first.

8. Customer enthusiasm is valuable evidence, but it is never a universal rule

Fokin’s largest decade-long process improvement is “massively bigger focus on the customer.” His painful 2015 Agrify mistake taught him that customers can value a product without becoming “raving fans,” a distinction he believes better customer work might have revealed before the psychological and P&L damage.
Customer research should surface the value proposition, reasons for staying or switching, competing solutions, purchasing process, and the decision-maker inside a B2B organization. Personal-network introductions may yield the least filtered answers, but cold outreach and expert-call providers extend the reachable universe.
Walker’s pushback—worth keeping—is that customers can be fickle about future products. A buyer may insist an upgrade is unnecessary, then test the launched version two weeks later and replace the entire installed base, making pre-launch stated intent dangerously fragile.
Fokin rejects Walker’s desire for “one rule to rule them all”: “There are no one rule answers at all.” Research, modeling, conferences, management meetings, web searches, AI, and customer calls merely feed the output; managers are ultimately paid for judgment and decision-making.

9. The right research tool changes with the question being asked

Fokin separates unknowable breakthrough demand from observable satisfaction with an established product. Steve Jobs-style reasoning concerns a B2C product and true breakthrough innovation; predicting the iPod’s reception was fundamentally different from interviewing 20 existing HubSpot customers about a functioning B2B platform.
A true breakthrough may remain “unknown and unknowable,” and correctly foreseeing it could be enormously valuable. Mature-product interviews, by contrast, can help establish whether users are satisfied, enthusiastic, switching, or expanding—evidence with a very different reliability profile.
His accidental but effective golf analogy lands the process lesson: golfers carry multiple clubs because terrain and conditions vary. Walker keeps seeking a single club; Fokin argues the craft consists partly of knowing which research instrument fits each situation.

10. Borrowed ideas deserve faster attention, not cheaper underwriting

Walker worries that a small-cap community can converge on the same names through letters, shared diligence, and mutual respect, much as clusters of larger hedge funds or Tiger-related managers sometimes do. The danger is outsourcing thought while mistaking social reinforcement for independent conviction.
Fokin’s one-line rule is “make the conviction my own.” An idea from a trusted peer can leap to the top of the research pipeline, but it should not enter the portfolio until it survives the same bar as a self-generated idea.
Routines enforce that independence. If the normal process requires three years of calls and conference presentations, five customer expert calls, or management questions, none should be skipped because a respected friend already performed them; Fokin concedes that, being fallible, he may not always meet his own aspiration.
Source calibration matters as much as source quality. Fokin wants close attention when he brings an obscure $200 million company with no coverage or expert calls; if he pitches a coal company at 4x earnings with a special dividend, Walker should “hang up” because that setup lies outside the relevant pattern.

11. Portfolio ranking must incorporate both expected return and possible ruin

Fokin compares investing with his former life as an international tax lawyer: a memorandum that was only 60% correct could deserve expulsion from the 38th floor, while being right 60% of the time in portfolio force-ranking may be excellent. Poker players may possess a psychological edge because they accept that a strong hand can lose.
He has become more systematic about comparing expected IRRs, but Walker finds raw rankings gravitate toward highly levered companies because equity upside is mechanically amplified. Charging a risk factor is difficult, while ranking on enterprise value can instead favor cash-rich businesses whose apparent discount offers little upside torque.
Fokin proposes at least two variables: expected IRR and a substantial penalty for actual business failure, not share-price volatility. If two investments each show 25% expected IRR over three years, one debt-free and the other described as having 6x EBITDA, 4x debt, and 2x equity, “these two opportunities are not born equal.”
Some risks are small in probability but existential in consequence. Fokin frames this as perhaps a 3% probability of something very negative, such as a product being outlawed nationwide, or a coup that transfers a foreign copper mine to a new regime. Walker adds China’s Alibaba/Ant episode as the kind of government-intervention risk he cannot confidently price; neither pretends an exact adjustment exists.

Evolution designed us to die fast; we can change that — Jacob Kimmel Can AI Fix Housing and Healthcare Affordability?