The Supply Chain of Talent

Plus! Getting the Cycle Right; Regulatory Arbitrage in E-Commerce; Dry Powder and Meta-Stability; The Split Market; Transmission Mechanisms

In this issue:

The Diff July 1st 2024

The Supply Chain of Talent

There are two main reasons The Diff is a recruiting service and early-stage investing syndicate in addition to being a newsletter. One of these is that money is nice, especially when there's a way for the same work to be part of more than one job—you can create a synthetic hundred-hour workweek if you're putting in more modest hours but some of those hours are contributing to two or three jobs at once.[1] The other reason is that if you work in an industry, and then start writing about that industry, it's very easy to anchor that writing to how the industry worked when you left. There isn't a great long-term solution to this other than tolerating some level of experience decay and trying to manage around it, but the next best thing is to have daily conversations with people who are still in the trenches, and to pay particular attention to what's changing.

One such change I've noticed: I started recruiting actively in 2021, which was a great time to be a) getting paid to find engineering talent, and b) getting paid a percentage of what those engineers were making. 2022 was not like that, to put it mildly. And then, in 2023, companies showed up with newly-reset plans, realistic growth expectations, and roles that needed filling. But this round, there was a big difference: much, much less willingness to hire junior developers. That's anecdotal, but it pairs nicely with an anecdote I’ve heard a lot lately: people who have a few impressive internships, and a degree in a technical field from a reputable school, are having a surprisingly hard time landing that first entry-level developer job.

But talking more about this problem raised another question: why were there so many of these jobs in the first place? Engineers are expensive, and that means training has a high opportunity cost. If the existing engineers are training the newest team members, that's a big investment that comes at the expense of whatever else they could be working on.

In many fields, there's a tacit division of labor: schools do the initial selecting, teach students theory, and give them a chance to network that they may or may not take. Their first real-world job is their first mandatory dose of practical training.[2] That makes sense in industries where the ground truth changes fast; schools know that frameworks and languages come in and out of popularity too quickly for a typical syllabus to keep up, while the knowledge from a class on compilers or discrete math will matter for longer. A junior investment banker modeling a tuck-in acquisition is not going to spend much time using the theory of uncovered interest rate parity. But in that case, part of the point of those jobs is to give people specific training that will make them economically valuable to future employers, and then monetizing some of that skill while they're still in-house and continuing to benefit from it later on—being a sell-side analyst who hires and trains entry-level analysts is a great way to make sure that in a few years, your social circle has lots of hedge fund and PE analysts.

That pattern doesn't seem to exist in tech, where there's already a norm of high turnover. Instead, companies seem to tolerate the fact that they're making entry-level employees far more valuable and then seeing them walk to somewhere else. That's a stable norm, though there are tweaks around the edges—waiting a year to vest, having staggered vesting, or otherwise deferring compensation can make the retroactive cost of hiring someone lower if they quit and higher if they stick around, which is a trade many companies are willing to make. The very biggest and very smallest tech companies have additional tools to boost retention:

Companies in the middle are in a tougher position: they'll be using more technologies that cleanly transfer to other jobs, but they're big enough that employee comp is more heavily weighted to cash and liquid equity, so there's less friction from switching jobs. The biggest companies have slimmed down a bit, and because they overhired in 2020-21 and then overestimated attrition in 2022, they've wound up with a decent surplus of employees who have a few years of experience and don't have much pressure to rapidly grow headcount. Meanwhile, rounds are smaller than they used to be, and companies are more judicious about hiring—if you're going from twenty employees to forty over the next year year, it's perfectly reasonable to accept some variance. But if the increase in headcount is more like five people, it makes sense to make lower-risk hires. (It also means that the existing team will spend less time interviewing candidates and answering questions once they start work.)

The nature of the work has also changed. The classic kind of task junior developers get assigned is something that's straightforward to describe, but a bit of a pain to actually do, and then straightforward to evaluate when it's done. The going rate for a developer who can do this is $20-40/month in the form of Claude/ChatGPT or Github Copilot.

There are three ways to hit a sustainable equilibrium:

  1. A wider gap between first- and third-plus-year compensation for computer science majors, driven mostly by a drop in first-year comp.
  2. A shift towards real-world skills in schools. This leads to a tradeoff, though: in both the finance and CS examples, the biggest companies have organized themselves around the need to get a large number of twenty-two-year-olds up to speed on their jobs. So this would have negative signaling value: it's essentially a way for people with top-tier credentials to increase their odds of getting non-top-tier offers, and of course employers will know this. There are already trade schools for programmers, and Excel/modeling prep courses for finance people, so it's also hard to see this as a new offering rather than a redesigned bundle.
  3. A change in the nature of compensation, where more entry-level employees are paid disproportionately in equity or have some of the cash component of their pay held back.

In the long run, one feature of growing industries is that the biggest pay premium they pay in absolute terms is to more senior people, who have developed plenty of unique skills, equity, personal connections, and have survived the inevitable career attrition of high-growth fields. But the biggest relative pay increase they offer—the place where there's the biggest percentage gap in compensation purely based on chosen career track—happens at the start. It's easy to overpay if you're confident that it positions you to be the high bidder for talent later on, when there's more certainty about what they're worth. Those early-career employees are a cheap call option on talent that's expensive to acquire later. When hiring slows down, that variance compresses, and the option value of the early hires takes a big hit. Adjusting takes time, for employers and employees, and sustained AI-driven growth in demand for developers might defer it. But that's eventually the direction such jobs go in: they still pay well, but part of what employers were paying for is an option they no longer need.

  1. It's analogous to the quantitative approach of having multiple signals that don't earn a profit after transaction costs, but which can be profitable in total if the same trade happens because three different signals say it should. The same hour of research can be meant to better understand a role someone's recruiting for, figure out the dynamics in an industry where a promising company is raising funds, and then turn into a newsletter issue. ↩︎

  2. In tech and finance, there are plenty of people who get hired without having done directly relevant work, but the best hires tend to be people with hobbies. They're just different hobbies. Launching a failed app and losing all of your savings trading soybean futures are two sides of the same coin—they're both demonstrations of an interest in the practicalities of the job, not just the knowledge taught in schools. ↩︎

Diff Jobs

Companies in the Diff network are actively looking for talent. See a sampling of current open roles below:

If you’re at a company that's looking for talent, we should talk! Diff Jobs works with companies across fintech, hard tech, consumer software, enterprise software, and other areas—any company where finding unusually effective people is a top priority.


Getting the Cycle Right

In late 2020 and early 2021, lumber prices roughly tripled as a rates-driven homebuilding boom coincided with a long period of underinvestment in sawmills. This ended up feeding into a supply-side inflation debate—maybe the impact of high deficit spending and zero rates would have been smaller in a world where companies spent more on capex and less on dividends and buybacks. But those recalcitrant reinvestors were ultimately vindicated: lumber prices are back to where they were in May 2020, and the companies that invested in more sawmill capacity aren't monetizing that investment well ($, WSJ). There are two ways to describe the art of running a company in a cyclical industry: survive the downturns, and, during upturns, make sure the profits don't all get sunk into buying new assets at cyclically high prices. Solving the second problem means the first takes care of itself—fewer big investments means lower fixed costs when prices inevitably come down again. And given the usual way industry demographics shake out, very few people entered the home construction industry after 2006 or so, and the ones who did had a hard time getting promoted. Which means that most of the companies in that industry are run by cautious survivors, not avid risk-takers. In this case, they passed on taking what turned out to be a pretty dumb risk.

Regulatory Arbitrage in E-Commerce

One bear case on Amazon is that they can't match the price offered by Shein and Temu, which take advantage of a trade loophole that eliminates tariffs on shipments worth under $800 ($, Economist). Amazon plans to use exactly the same loophole for its own deep-discount service ($, The Information). Amazon is in an annoyingly tricky position here: for a long time, they were able to treat fast shipping and low prices as complementary objectives, both of which benefited from having more Prime subscribers, a denser fulfillment network, and better tools for predicting and managing demand. But this made them gradually worse at serving the needs of customers who were willing to wait a week or two in exchange for saving a few dollars on a transaction. To a company that pays close attention to customer-level profitability metrics—and Amazon was quite early to emphasizing a subscription/membership model compared to other online retailers—these price-sensitive customers were not the ideal market to target. But there aren't many consumers left who haven't at least considered Amazon, and whether those new customers end up being Prime subscribers in a few years or not depends heavily on whether Amazon is cheaper than Temu for the next purchase they plan to make.

Disclosure: Long AMZN.

Dry Powder and Meta-Stability

The Bank of International Settlements has the enviable job of publishing interesting research, holding conferences, and sometimes lecturing central bankers. They're doing a bit of option three by suggesting that central banks be cautious in cutting rates ($, FT). In most countries, there's a domestic case for cutting rates—slow growth, declining inflation—and a global case for not doing so, because the US economy is so strong that rates have remained higher here, pushing the dollar up. Large, expanding rate differentials can make sense in a local context, but will also mean that more money flows out of other countries and into the US, which could ultimately result in financial crises in developing markets. But this also illustrates one of the tricky incentives central banks face: it's easier for them to use standard policy tools, like rates, than to invent new techniques as they had to do in 2008 and in 2020. They have more room to do this if rates are a bit higher than they'd naturally be. They're more likely to get blamed for one month of a crisis than for years of slightly below-potential economic growth, so they have a moderate technocratic incentive to keep rates slightly high.

The Split Market

The WSJ has a good recap of last quarter ($), noting that the market is flat outside of the outperformance of AI-related stocks, and that most sectors are down. Any time you slice up the aggregate change of some index whose components can be up or down, you can tell a similar story—there's usually some sector that's driving disproportionate growth and getting disproportionate investor focus, and the rest of the market looks worse for it. Since bull markets usually have a narrative, and that narrative is tied to whatever sector is doing best, this also means that the better the market has done, the easier it is to tell a story about the economy's wholesale dependence on a single sector. That can run into risks—housing, and home equity withdrawals, were a big contributor to economic growth in the early 2000s, and we know how that turned out. But this kind of observation will always be possible.

Transmission Mechanisms

One place where it is very useful to think about constituents rather than aggregates is in measuring the impact of some policy effort to increase consumer spending. In China right now, the central government is encouraging local governments to subsidize consumption, but those local governments are reluctant to do so because they directly bear the cost of that spending, and they're already in poor financial condition. Japan in the 90s and much of the world in the 2010s showed that if you provide additional liquidity to a sufficiently indebted sector of the economy, the result is that that sector deleverages rather than increasing its spending—the impact of that spending on growth is minimal. (The flipside of this is that the impact on inflation is also minimal: both the US and Japan in those periods had high deficits that didn't spark inflation because they basically represented a gradual shift of leverage from corporate balance sheets to the government's balance sheet.) The Diff has argued before ($) that China would have more fiscal flexibility, and a better shot at being a reserve currency issuer of sorts, if more of its debts were on the balance sheet of the central government rather than off-balance-sheet credit provided to more local political entities. If their government keeps trying to stimulate consumer spending, and keeps running into the same problem, that's a course of action they're more likely to take.