# The Sentence He Didn't Say: Notes on Speed, Dissent, and the Job Losses AI Hasn't Caused Yet > Personal notes after listening to Bloomberg Odd Lots (2026-09-01) with Adam Posen: the difference between direction and speed in central bank communication, why committees drift into silent agreement, and why AI's labour market effects still aren't in the data. Educational reflections, not investment advice. Published: 2026-09-01 Locale: en Tags: Federal Reserve, inflation, central bank communication, AI and jobs, decision discipline ![An empty mountain lecture hall at dawn, low light through tall windows, an open set of speech notes left on the lectern, rows of chairs receding into depth](/covers/oddlots-2026-09-01-adam-posen-thinks-things-could-get-very-messy-for--cover.png) > When the sage makes laws, he must make them plain and easy to know, with names set straight, so that the foolish and the wise alike can all understand them. > > —— *The Book of Lord Shang*, "Fixing Rights and Duties" (Warring States period; my own translation) That line is over two thousand years old and it's about something unglamorous: a rule isn't a rule until someone slow can follow it. I listened to Bloomberg's Odd Lots from 1 September 2026 — Joe Weisenthal and Tracy Alloway at Jackson Hole with Adam Posen, president of the Peterson Institute — and that sentence kept coming back to me. They're discussing the most consequential committee on earth, and the central problem is just as unglamorous: was it said clearly, or does the rule live only inside one person's head? ## What the episode is about Posen served on the Bank of England's Monetary Policy Committee and doesn't varnish much; the hosts open by saying that's exactly why they invited him. The conversation runs in three parts: what the new Fed chair's Jackson Hole speech did and didn't say, the unwritten rules inside central bank committees, and then a turn toward AI — economists leaving academia for AI labs, and why the AI job apocalypse still isn't showing up in the data. The part I got most from wasn't his rate call. It was the point he kept circling back to: the dangerous state for a decision-maker isn't being wrong. It's being unpredictable about *how* they will decide. ## The main points **1. A B-minus, and the sentence that never arrived.** Posen grades the speech a B-minus by normal standards — better than that in context, given how confusing the preceding months had been. His sharpest observation: the fourth section lays out the whole litany of reasons inflation might persist or drift higher. Logically, the next sentence is "and therefore we need to tighten." That sentence isn't there. That's the "messy" part — if they then don't hike, people start asking whether it was political pressure, or whether the chair got out over his skis ahead of the committee. Listing every reason and withholding the conclusion just hands the cliffhanger to someone else to fill in. **2. His real worry isn't hawkish or dovish — it's discretion preserved until the last minute.** The speech closed by saying inflation must move in the right direction at the right speed. Tracy pushed back on the spot: what's the right speed? Sounds awfully subjective. Posen went further — without a speed, the direction means nothing. Inflation has been above target for years; everyone knows which way it's supposed to go. If you never declare how fast you'll get back, "the right direction" carries no content. That's the line I most wanted to keep. **3. The "Weekend at Bernie's" problem, and where inflation targeting came from.** Central banking has always situated itself somewhere between rules and discretion. Pure discretion peaked with Greenspan around 1999 — he could get away with it partly because he was very good and partly because he dominated the committee. Posen's vivid worry: if an already-elderly Greenspan had had a heart attack on the Fed's tennis court, the credibility would have evaporated, because all of it was attached to one man. That's precisely why Bernanke, Mishkin, Laubach and Posen pushed inflation targeting and accountability — not to handcuff anyone, but so the institution wouldn't live inside a single skull. **4. The unwritten rules bind harder than the written ones.** This was the part that gave me a chill. The ethic in the Fed system is that you may dissent but you don't contradict the chair — and, as former governor Larry Meyer pointed out, there's an informal ceiling on how many people dissent at any one meeting. The chair never loses a vote. Even Volcker, once it was clear he was going to lose one, said the next meeting would be his last. Worse is the momentum: Posen says that after twelve or fifteen meetings with no dissent, the psychological bar for the next person to dissent rises on its own. He thinks the low dissent rate under Powell was partly deliberate and partly defensible — during COVID, and while the Fed was under open attack, not showing division made sense. The cost is that the bar got raised. **5. Where political interference actually begins.** Posen's distinction is clean: a certain amount of yelling at the central bank is part of the game, and historically it was sometimes healthy — the president complains loudly in public while, in the background, winking at the Fed to do its job. Two things are different in kind. One is caving under fiscal pressure; the other is going after personnel, votes, and budget — using appointment timing to hollow out functional independence. The best line of the episode lives here: when a Treasury secretary or president says *you have to help me sell US bonds*, the right answer is — I'm here to help you sell bonds, and to help your successor sell bonds, and your successor's successor; which is exactly why I don't cheat now. Turning long-term credibility into a reason to say no is a frame that travels well beyond central banking. **6. The likeliest surprise is the communications committee.** Everyone's watching the balance sheet group, which Posen guesses will come back more sober and small-scale than expected. The radical one, he thinks, is the communications task force led by Mervyn King. The role reversal is the fun part: King led the inflation targeting push in 1992, when sterling crashed out of the exchange rate mechanism and Britain needed an anchor, and he fathered the fan charts — here's our central case, here's the range it might land in. Tracy says she loved the fan charts; Posen says a scattered few did, but it was a niche product. Since retiring, King has become loudly sceptical: central banks forecast badly, publishing all that shaky information is noise, and markets grow too dependent on guidance and stop pricing risk properly. In the speech, the chair conceded a "hall of mirrors" — if the Fed chases market noise you get a strange dynamic. Posen was glad to hear it, and much preferred it to the earlier "play the ball, not the referee," which he calls the single most annoying thing said to the central banking community. **7. On AI: some sign of productivity, none in employment.** If you had to name two jobs you'd bet were toast, Posen says, you'd pick long-haul truckers and junior coders — and hiring in both is still growing. He points to Luis Garicano and co-authors on "messy jobs": almost every job is embedded in specific knowledge and relationships far denser than the "most exposed sectors" lists produced by consultancies and international bodies. His historical check: in nineteenth-century England the skilled hand weaver really was replaced by automated looms — that part was real — but had you run this kind of study before the Industrial Revolution, the displacement outside that narrow job description would have looked very different from what you'd have predicted. The funniest moment is his account of a closed-door meeting where a very senior figure from the AI world was asked why the job apocalypse hadn't arrived, and answered that it turns out people quite like dealing with humans and don't always want to deal with machines — at which even a room of not-notably-emotionally-intelligent economists thought: well, obviously. ## Going further ### "He said all of it — so should I act or not?" That's the honest position a lot of us are in after an official's remarks or an earnings call: every sentence was comprehensible, and the sum tells you nothing about what to do. I used to blame my own lack of expertise. Often the statement just wasn't finished. Posen's cut is portable: split any claim into direction and speed. Direction is "inflation will come down," "margins will improve," "the project will accelerate." Speed is "by when, to what level, and what happens if we miss." A direction-only claim survives every outcome — inflation falls and we called it right; it doesn't and it was an external shock. You can't mark it against reality, which means it gave you no new information. I do something mechanical with earnings calls now: next to every promise I leave a box for speed, and put an X in it when I can't fill it. If a whole call is X's, I file it as a sentiment signal rather than a fundamental one — not that anyone is lying, just that nothing said has entered a verifiable state yet. The proportion of X's turned out to be much higher than I'd assumed, uncomfortably so. The flip side is why direction-without-speed talk so often gets a first-day rally. Naming a speed is putting a collar on yourself, which hurts, which is why it's rare — and rare things deserve extra weight. ### "The news says AI will replace everything. What about the thing I'm holding?" The second half of the episode answers a more personal question. I've read plenty of "most exposed occupations" rankings; they make me uneasy and I've never known what decision to make with one. That's exactly Posen's objection: they use job titles as the unit, and work isn't shaped like a job title. Tracy tells a story I loved. She and her husband were roofing an outdoor shed, asked ChatGPT, watched a couple of YouTube videos, and used a recommended product for sticking down the shingles. Partway through they discovered it only works above 60°F — and they were in Connecticut, not Florida. Neither the bot nor the videos caught it, because nobody had told them where "here" was. The point isn't that AI is useless. It's that the unstated premises — today's temperature, the state of this particular building, what last year's repair left behind — are where the expertise actually lives. That's what "messy jobs" means: not that the skills are hard, but that the context is thick, and context is usually unwritten. My working version of this: to judge how secure a job or a company is, don't ask how copyable its skill list is. Ask how many facts it handles daily that aren't written down anywhere. The more of those, and the more they're tied to specific people and places, the harder it is to swallow whole. Conversely, if a company's value is mostly "knowing the standard procedure," that's an awkward place to stand. There's also a sense of timing here that I find more useful than the conclusion. Posen invokes Brynjolfsson's J-curve: after a technology arrives, there's a stretch where human-plus-AI is genuinely the efficient arrangement while firms slowly reorganise around it. He won't say whether that stretch is one year or five — probably under ten, probably nearer five. He also cautions against attributing today's weak hiring of young people to AI automatically; the post-COVID reshuffling of roles is still working through. What I take from this is to keep three things apart: "hasn't shown up yet," "won't happen," and "already happened but isn't in the numbers." Headlines routinely blend all three. His 1990s comparison is worth holding too: Intel and TI following Moore's Law showed up a bit in GDP, but the real gains arrived when McDonald's, Walmart and UPS rebuilt their businesses around it. That says something fairly direct about how to weigh the people selling shovels against the people learning to dig. ### "Twelve meetings with no dissent" — you have that state too This is the bit I've sat with longest. Fifteen meetings without a dissenting vote, and the bar for the next dissenter rises by itself. Nobody acted in bad faith. It simply grows. Posen is fair about it: the low dissent under Powell was partly reasonable, given COVID and an institution under open attack. He supported something similar himself at the Bank of England, when King said quantitative easing was genuinely controversial, take the vote however you like, but afterwards could we have a moratorium on relitigating whether it works. Posen backed that — it was an emergency. The trouble is emergencies end and habits don't end themselves. My own records have the same shape. If nobody — including me — has raised a bear case on a position for a few quarters, I check it less and less, on the grounds that it's never been a problem. But "never been a problem" isn't evidence; it's the residue of my not having looked. It has nothing to do with the fundamentals and everything to do with the momentum of the record itself. Posen's passage on economists moving to AI labs is the same illness from another angle. He's honest about it: most of them sincerely believe AI is the most transformative technology of the century, and the motive isn't dirty. But he says plainly that once you join a hierarchy you lose an independent voice, and what you choose to work on — and not work on — shifts with your position. Not outright corruption, but unsettling. His own hope is that some people will keep settling for low six figures and shape policy from outside. Reading those two passages together, what I come away with is this: judging what a claim is worth means looking at where the speaker is standing and how long it's been since anyone next to them said no. Neither has anything to do with intelligence or honesty, and both systematically change what you hear. ## Worth looking at - Bloomberg Odd Lots, 1 September 2026, "Adam Posen Thinks Things Could Get Very 'Messy' for the Fed," recorded at Jackson Hole. - Public research from the Peterson Institute — the episode cites Joe Gagnon and Martin Chorzempa on semiconductor output and GDP attribution, and Jed Kolko on the labour market. - Luis Garicano and co-authors on "messy jobs," and Erik Brynjolfsson on the J-curve, are the scaffolding behind the AI-and-employment discussion. - On the origins of inflation targeting and accountability, the Bernanke–Laubach–Mishkin–Posen book on inflation targeting, and the original design intent of the Bank of England fan charts. - For a direct comparison, FOMC minutes and voting records are public. Counting dissents per meeting tells you more than most commentary. ## The one thing to take away Just one idea: **direction isn't a commitment; speed is.** When anyone — us included — says a thing will get better, without "by when, to what level, and what if we miss," that sentence is true under every outcome, and therefore useful under none. It isn't a lie. It just hasn't become something you can mark against reality yet. Here's a thing I tried, simple and unexpectedly uncomfortable: take one thing you told someone (or yourself) in the past two weeks that you'd do — exercise, change jobs, have the conversation you've been avoiding, straighten out an account — and rewrite it with speed attached. Three slots: by when, what counts as done, and what happens next if it isn't. The first time I did this, half my sentences stalled at the second slot, because I couldn't say what counted as done. That stall is usually the place where I hadn't actually decided to do it.