Summary

Fennerty’s single biggest fix for unsatisfying expert calls: approach each one as “I am testing a hypothesis or a thesis and I want a thought partner who’s credible to think through that and the second-order implications.” The best calls have a goal, enough structure to test it, and enough flexibility to probe — “flying at the right altitude” — while investors hunting “data that corroborates this thing” can leave frustrated that the expert was evasive.
The economics of expert calls have been rebuilt in three to four years: the former $1,500–2,000 average price for a private call gave way to Tegus-style at-cost calls monetized through a searchable transcript library. The result: “the stuff that we used to spend two weeks just getting up to speed on, we do now in a day,” and live-call work now concentrates on two to three thesis drivers with three to ten credible experts. Private-market investors have spent the last 18 months mimicking what public investors did 18 months before.
Bias is inherent to every insight category — management guidance, sell-side, backward-looking financials — so the answer is triangulation, not elimination. Fennerty’s account of top investors’ toolkit: confirm the expert’s actual purview up front, deploy a “barometer question” (ask a bullish expert about culture and see whether the story exposes a problem), close with “let’s say we’re both wrong… what do you think we might have missed?”, and never rely on one operator view — “that’s a leap of faith.” On disgruntled formers, his reference-call parallel: “I have to do seven to eight references to really triangulate to truth.”
AI interviewers are collapsing the cost model for survey- and channel-check-style work that expert networks could never economically deliver. A single survey could run $100K versus $2K for a call; now an AI interviewer can talk to 10 CIOs “on their clock” — “stuff that was really hard to operationalize even six months ago.” Relatedly, “being a really expert notetaker on the back of your call has a very short half-life.”
Fennerty’s claimed superhuman edge for AI is cross-source synthesis, not investment judgment: think “an 80 by 80 grid comparison of inputs across multiple data sources” that a human analyst could not feasibly perform, or a prompt comparing CEO guidance against the cash-flow statements of the last five comps to flag a possible outlier or overconfident management. Portfolio monitoring is going “custom autonomous” — Friday reports on trends and inflections, as if you had “infinite analyst resources” — with Liberation Day cited as the fire-drill proof case.
Fennerty “believes in his bones” that by 2030 there will be high-performing PMs who never built a detailed M&A model — because “your technical prowess and your analytical skills… over time are getting commoditized” while pattern recognition and judgment become more important. But trust requires everything “fully traceable down to the source,” because “AI is very prone to, if you prompt it a certain way, it’ll pound the table.” The public-markets skill shift: “from the analyst skill set to the architect skill set,” possibly toward something “not quant investing, not fundamental, but something in between.”
The tradeable meta-claim: AI “has absolutely collapsed the resource advantage that the biggest funds have had” — a mid-market fund now effectively fields “a crack team of incoming KKR analysts” — while human “spidey sense” remains an important complement for catching frauds and bad recommendations. Walker’s counter-worry that scale advantages accrue upward via proprietary data goes unresolved; Fennerty’s close is that fundamental investing retains a role and the generalist-versus-specialist debate does not change, while “there’s a lot of hype, but there’s also a lot of real stuff happening.”

Deep dive

1. The one fix: treat the expert as a thought partner, not a data vending machine

Walker’s framing of the whole conversation: he does roughly 25 expert calls a year, “could be upwards of 50,” and grades them “10% awesome, 50% good, 40% okay, and 10% bad” — the goal is moving awesome to 30% and eliminating the bad tail.
Fennerty’s number-one takeaway, at the acknowledged risk of generalizing: frame calls as “I am testing a hypothesis or a thesis and I want a thought partner who’s credible to think through that and the second order implications.” The best calls have a goal plus structure with flexibility to probe — they’re “flying at the right altitude.”
The failure mode he describes: investors who “come in really trying to say I just want data that corroborates this thing I’m trying to test,” then come out frustrated that the expert was evasive or “gave ranges that didn’t make sense.”

2. Transcript libraries rebuilt the economics — and raised the bar for live calls

The old model, which Walker lived in consulting and PE: “$1,500 to 2,000 was the average price for a call,” all private, deployed in 8-to-40-call diligence sprints. Tegus monetized through a searchable transcript library instead, doing calls essentially at cost — Fennerty’s analogy is Spirit Airlines expanding who could fly, with much of the business built on mid-market funds that previously couldn’t afford the volume.
The consequence: first-order ramp-up work — market structure, go-to-market, pricing, operating leverage — moved to the libraries. “The stuff that we used to spend two weeks just getting up to speed on, we do now in a day.” The second, third, or fourth call becomes the first.
Where live-call effort goes now: pick two to three drivers central to the thesis, then “get three to 10 credible experts, really dig into that and validate that.” Adoption arc worth noting: private-market investors have spent the last 18 months mimicking what public investors did 18 months before.

3. Walker’s echo-chamber worry vs. Fennerty’s N-count answer

Walker’s concern: if five funds drive ten calls on a cultish tech name in August 2025 and everyone reads the same transcripts, “everybody’s thinking about and coming at the company the same way.” Fennerty’s candid response: “we haven’t heard that as a concern.”
His reframe: expert insight is inherently biased, like every category — “management guidance has a bias, sell-side research has a bias… financial data is backward-looking.” The bigger bias to interrogate is the one embedded in the individual expert’s experience, rather than assuming the investor-created library eliminates bias.
The load-bearing discipline: “the way to avoid bias with operators is to go get multiple operator views. You don’t need 30, but relying on one operator view to really prove or disprove a thesis is obviously dangerous. That’s a leap of faith.”

4. In-call craft: the barometer question and the closing curveball

Three habits of the heaviest, best users. First, an immediate double-click on where the expert sat and what they could actually see — “the lens from which this person is coming from, what they saw and what they couldn’t see.”
Second, the “barometer question” — a mid-call gut check on tonality. The example as told: an expert is bullish on the business model all call, then you ask “talk to me about the culture. How has that shifted?” and suddenly “actually there’s a really deep problem there… the culture’s gotten a lot worse recently” — an instant hint of internal misalignment worth pulling on.
Third, the open-ended close: “let’s say we’re both wrong on what we just discussed… What do you think we might have missed? What could go wrong?” Fennerty’s claim: experts are uniquely good at second- and third-order risks invisible from outside — which matches Walker’s experience of specialists surfacing risks “they live and breathe” that he’d literally never considered.

5. Formers are negatively biased by construction — calibrate like a reference check

Walker’s structural point: most experts are either competitors (the Pepsi employee on Coke) or formers — and people are usually formers because of layoffs or getting passed over, so the pool skews disgruntled. He also confesses his own tell: a bullish expert “knows what he’s talking about,” a bearish one is “a clown.”
Fennerty’s calibration method: know the bias exists and that risk may be overstated, then use spread across multiple formers — three of three saying broadly similar things is probably credible; three of three negative with “varied levels of tonality” supports a different assessment. His reference-call parallel: “I have to do seven to eight references to really triangulate to truth. Every time I do that, I get one or two that had I taken them at face value would have really colored the picture very deeply.”
On screening from thousands of daily projects: the best outcomes come when investors specify who they want and why; the failure mode is “we want to talk to people with this title and that’s the amount of context.” Key distinction: “Seniority is not the same” as flying at the right altitude — senior titles are often too disconnected from operating-level questions on inventory or supply chain.

6. AI interviewers crack open surveys — and make expert note-taking much less central

Fennerty distinguishes deeper business-model calls, which are generally more satisfying as longer, in-depth conversations, from real-time market-pulse calls — both work — but survey/channel-check-style signal collection has been “frustrating or unreliable” and cost-prohibitive: “you’re not going to spend 100,000 for a single survey whereas you could spend 2,000 for an expert call.”
His early-but-big claim: AI makes that cost and operating model “vastly different from what we’ve ever experienced in the industry” — an AI interviewer can talk to 10 CIOs “on their clock,” getting real-time insight that “was really hard to operationalize even six months ago.”
On Walker’s note-taking struggle across four transcripts on a name over six months: the discipline of post-call synthesis persists at good funds, but “being a really expert notetaker on the back of your call has a very short half-life” — instant transcription plus AI synthesis in your preferred structure is where “within months most people are going to be moving.”

7. What AI can do superhumanly: 80-by-80 grids and always-on monitoring

The earnings-season use case: pick where you do “hand-to-hand comparison and synthesis” under time pressure. In the Reddit example, a real investor’s prompt compares what the CEO is saying against “the actual cash flow statements of the last five comps that I tell you have already reported” — the answer is either Reddit is an outlier or “management’s overconfident, and we’re already setting up for a question mark.”
Walker’s pushback — worth keeping: that sounds like pod shops trading quarters and whisper numbers; what about the five-stock concentrated long-term investor? Fennerty’s answer: differentiated view versus consensus, built by comparing management guidance, sell-side debates, your own expert calls, library transcripts, and internal views — “an 80 by 80 grid comparison of inputs across multiple data sources is just not feasible for a human analyst to do,” but it reveals the real debates worth deeper work. Walker’s corollary: nobody reads 80 transcripts a year on each of 30 names; AI can.
Portfolio monitoring has gone from search, to workflows, to “custom autonomous things that run reports as if I had an analyst working on it” — Liberation Day as the fire-drill proof (exposure and research that differed from or aligned with the investor’s view identified within hours), and the steady state: “every Friday I want a report in this format that tells me trends and inflections… against my portfolio. Imagine if you had infinite analyst resources.”

8. The hand-built-model debate: judgment matters more, technical prowess commoditizes

Walker’s confession: he builds all his models by hand near a decision because doing the work is how he internalizes it — and worries AI summaries mean “I don’t think it through as much… I’m just outsourcing it all to AI.”
Fennerty said he’d felt the trade-off himself, but: “I believe in my bones that by 2030 there are going to be really high performing portfolio managers… who absolutely never had to go through that” — never built a super-detailed M&A model, yet get good outcomes.
Two trust conditions from his own use: everything must be “fully traceable down to the source” (“I don’t want to get two hours in and suddenly have it all be on a shaky foundation”), and beware that “AI is very prone to, if you prompt it a certain way, it’ll pound the table” — his own go-to-market plan example, where customer verbatims and TAM judgment overrode the model’s conviction.
The reallocation he draws: “your sense of self as an investor is your technical prowess and your analytical skills — those over time are getting commoditized, and what’s much more important is your pattern recognition, judgment, ability to push on these things.”

9. Conviction, not volume: PE compressed A-to-B from weeks to a day

Fennerty’s baseline-shift frame: like the PC and Excel, sophisticated analysis stops being differentiation and becomes table stakes; alpha comes from systems that let you “make decisions much faster with conviction.”
The PE specimen, with parallels for concentrated public investors: funds still do three to five deals a year, but “the time to get through point A to B in our process has compressed to a day from weeks,” so more time and energy goes into the B-to-C investment-committee debates on value creation and differentiated drivers.
Walker’s probe: if conviction is higher, shouldn’t it be one to three deals instead of three to five? Fennerty hasn’t seen that — some funds do more, some the same-but-more-convicted — but the funnel top has expanded: “some have said I’ve looked at twice as many things now,” triaging CIMs “green yellow red in a way that took weeks of analyst capacity.” The driver: elevated valuations and speed — “we can feel it around us how quickly people are moving on opportunities with conviction.”

10. The next alpha skill: architect over analyst — or “something in between”

Walker’s historical arc: 60 years ago you could win as a quant in your head (Ben Graham calculating net working capital); the last 10 to 15 years rewarded the qualitative call (Google, Facebook, Amazon as the best businesses ever); AI is now raising the qualitative bar too — with his Buffett caveat that when everyone stands on tiptoes at the parade, no one sees better. His own galaxy-brain bet for privates: skill with “human resources and people” gets elevated.
Fennerty on privates: returns from financial structuring and dealmaking have been fading, so gains shift to acting fast on a bigger opportunity set and to portfolio value creation post-deal — “all the buzz” at a mid-market PE conference was taking AI into portfolio companies to change operating models and cost structures.
On publics: “this shift from the analyst skill set to the architect skill set,” with pod-shop-style pressure spreading through the industry. His long-term question mark, left genuinely open: does a new cohort leapfrog the old pattern-recognition — testing “truisms that we’ve all lived with that are uncorrelated in the data” — producing something that’s “not quant investing, it’s not fundamental, but something in between”? Walker: “I am already a dinosaur.”
On Walker’s three-bias stack (self-bias in prompts, promotional bias in company documents, negative bias in expert training data), Fennerty’s honest non-answer: don’t oversolve for eliminating bias — “the triangulation you can do across these different sources and perspectives is infinitely higher than you could before, and that’s ultimately great investment work.”

11. The fraud knuckleball — and why AI may level the field without replacing spidey sense

Walker’s knuckleball, via the 2011–2014 Chinese reverse-merger frauds (“600 million acres of woodland” that didn’t exist): does AI raise the returns to fraud once quantitative screens replace the human who flies out and finds the $4 billion company headquartered “in the third floor of a mall”?
Fennerty’s rebuttal: nothing inherent makes AI a fraud accelerant — it’s already a fraud-detection weapon, e.g., satellite imagery of shipping lanes showing “the traffic is not even remotely what company guidance is.” What AI can’t do is anything “that remotely looks like an investment recommendation” at PM level.
His underdog thesis: AI “has absolutely collapsed the resource advantage that the biggest funds have had versus your mid-market funds” — like fielding “a crack team of incoming KKR analysts” — while the human whose “spidey sense goes, ‘This doesn’t make sense. I got to go dig deeper’” remains an important complement. Walker’s unresolved counter: scale might still win via literally proprietary data — analysts at every industry conference feeding internal libraries alongside the crossover funds already blending their own call archives with the public library. Side note: nearly 40% of AlphaSense’s business is built on corp dev, corp strat, and IR teams.
The close: Fennerty says fundamental investing retains a role and does not expect the generalist-versus-specialist debate to change. He also acknowledges big-fund advantages alongside smaller funds that may outcompete them, and the real task is sorting “what is table stakes to not fall behind and what’s true advantage. There’s a lot of hype, but there’s also a lot of real stuff happening.”

Klarna CEO: SaaS is Dead: Why Systems of Record Will Die in an Agentic World Jason Fried: Your Only Competition Is Your Costs