Deriving the Worth of those Claude Code Essays (and All Different Essays) utilizing Claude Code Integration to Substack

Substack built-in Claude Code not too long ago by way of an mcp (undecided if I used all of that accurately). You may examine that right here.

If you wish to join Claude Code to your substack, right here’s the directions to take action. However let me lay out what I’ve discovered, and the way I’m going to study from what I discovered going ahead. I had Fable 5 do the evaluation for this conversationally and I’ll attempt to hold it to a naked minimal. The aim, although, is to determine what’s the social worth of those Claude Code essays?

Historic Developments

I began this collection again in December thirteenth of final yr. The primary entry was known as “Claude Code Modified How I Work (Half 1)”. And since then I’ve written round 55 essays whole. I assumed it was 56, however it’s apparently extra like 55 with a ton of unpublished drafts in my drafts folder.

Since then, the substack grew a ton. There’s a very clear. Main as much as the day earlier than I began the Claude Code essays, I used to be on a linear path that reached 20,000 whole subscribers (paying and non-paying). However the second I began that essay, the substack exploded in reputation. Extra not too long ago it’s gone again to the identical linear progress, although, kind of. However as you possibly can see right here, it grew from 20,284 on December twelfth to twenty-eight,099 subscribers as of this morning, July eighth 2026. So, over 7 months, I believe I really feel fairly good saying that the Claude Code collection surprisingly added round 8,000 subscribers.

How lengthy had it taken so as to add 8,000 subscribers earlier than then? Round two years. You may see that November 4th, 2023 I had 12,000 subscribers. So from November 4th 2023 to December twelfth 2025, the Substack’s subscriber base had been 8,000 over two years. In order that alone was attention-grabbing.

Submit-Claude Taxonomy

I numbered 56 Claude Code essays whole, however in actuality there are actually 55 as a result of I deleted one about p-hacking. I requested Fable 5 to create a “style taxonomy” of them. There are 5 distinct kinds of substack essays on Claude Code I’ve been writing. They’re:

1. Financial essays (19 posts). The Financial Essays are argument-driven items about what AI does to analysis, publishing, and the occupation: the “Modified How I Work” manifestos (Elements 1–2, 4, 5), Zero Revenue Situation, Verification Is the New Bottleneck, Falling Marginal Prices, A Modest Proposal for Editors. Inside this style sits a definite sub-form I used to put in writing in that I known as “Claude Code fan fiction” speculative essays (CC27 Analysis and Publishing Are Now Two Totally different Issues, CC30 the Lament), plus the confessional (CC36 mea culpa on p-hacking). I particularly loved the fan fiction essays to be sincere. However I additionally get pleasure from writing the financial essays extra usually. Somewhat intermediate microeconomics goes a great distance in eager about the affect that Claude Code is having on empirical work and analysis staff.

2. Workflow tutorials (16) — A second fashion of essay are issues that depart the reader with issues they’ll copy. Form of DIY or “the right way to” essays that may be instantly used. These have been my MixtapeTools (P12), Abilities and Break up-PDF (P13), the deck immediate (P11), Dispatch, the continuous-DiD studying collection, harness rules. The concept of those essays was very sensible: depart the reader with one thing they’ll run on their facet. Not “huge suppose” essays like the primary ones. Simply narrowly about work product.

3. Experiment/audit logs (13) — Third one was my experiments and audits. These have been a collection of lab pocket book fashion essays with outcomes. They mainly “wrote up” outcomes from work I used to be doing utilizing Claude Code for empirical analysis as I used to be studying my approach round. Issues like my PNAS replication arc (P14, 15, 18, 19, 22), the multi-agent Callaway–Sant’Anna audit arc (24, 26, 28), the six-package apple-to-apple audit (CC31), the csdid guidelines saga. These have been additionally plenty of enjoyable, as these have been actually those the place I began to study extra about utilizing AI Brokers for sensible analysis functions by doing analysis right here. I took on issues that I actually couldn’t have completed very simply.

4. Subject notes (4) — diary entries: the four-talks-in-two-weeks recap, the Goldsmith-Pinkham plug, Amnesia.

5. Artifact showcases (3) — the factor itself is the publish: Kennedy Faculty video, lovely decks (P7), the faculty-adoption deck.

Claude stated that these essays even have “private hooks”, often early on, like the place I’ll discuss rising up in Brookhaven, Mississippi, having ADHD “primarily inattentive” subtype, or issues about Madrid. I knew I habitually discuss private stuff, as that’s a attribute I appear to fall into often, however I didn’t fairly understand I used to be doing it within the Claude Code collection too. It sounds prefer it simply creeps into my complete fashion of writing, although.

You may see right here how the 5 genres have completed. The essays win outright. There have been round 98 subscriptions per posts. It’s about 2.5 instances extra impactful than these subject notes measured by subscriptions per 10,000 views. However there are two outliers in it — essays 1 and a couple of — which I’ll have a look at it in a second individually, so bear that in thoughts. However general, essays led to 1,861 new subscribers.

Tutorial have been the second hottest, and on down. So that’s the place you see it. You get to round 3,500 new subscriptions whole from the collection measurable, despite the fact that there’s round 8,000 new subscriptions for the reason that day earlier than I began. But it surely’s actually “essays and tutorials”. That seems to what has been traditionally helpful for readers comparatively talking.

The primary two essays have been completely different although.You may see that breakdown right here. The very first essay had nearly 25,000 views with an open charge of 41.5%. Open charge is “what % of emails received opened”, and often my open charge is round 30%. I’m not fully certain how substack measures time on publish, or how readers completed, however it’s within the metadata curiously, and Claude grabbed it.

In order that’s kind of the place issues stand on the Claude Code collection, and I’ll hold writing about it as a result of I’m going to be persevering with to pressure myself to construct up a physique of fabric that helps me do empirical analysis utilizing Claude Code — which I can be posting about right here often — and I’m additionally going to be investing closely in integrating Claude Code and AI Agentic assisted empirical analysis into my lessons at Harvard Kennedy Faculty this coming spring 2027. I met yesterday with somebody at Kennedy to go over it, and I’ve their help to do it, and so subsequently I can be doing it full on. I’m instructing a full semester course on causal inference to the PhD college students, after which two half semester programs on causal inference (one to the worldwide/improvement PhD college students, and one to a different one I’ve now forgotten the precise title for). Kennedy’s help on this matter is an actual honor, I ought to say, for me personally. I walked across the constructing yesterday for the primary time and was simply sincerely each humbled and excited. I believe that it’s a good match for me personally, and I hope for them too, and I can be investing closely in attempting to equip PhD college students with each deep data about econometrics and causal inference, but in addition very utilized data which incorporates “doing analysis”, but in addition “utilizing AI Brokers” for that function. Anyway, the purpose I’m making is I’ll proceed to make use of the substack to doc my progress as I construct out a strong pedagogy round instructing econometrics and causal inference on the PhD stage that integrates AI Brokers in a approach that furthers my objectives extra usually whereas addressing lots of the pitfalls that I believe await all of us. So want me luck there.

Non-AI Style Taxonomy

However lots of people come to my substack for causes that aren’t associated to AI. Or relatively, some come for the AI, however keep for different issues, and a few come for the opposite issues and keep for the AI, some neither, some each. Anyway, what are these essays? I’m not going to hyperlink to those because it takes too lengthy, however I’ll break it up into two classes.

1. A complete essay style taxonomy. What are the kinds of essays I’ve traditionally written and what are they like by way of measurable helpful to the neighborhood?

2. Distinction-in-differences taxonomy. What are the kinds of sub-genre essays, for differences-in-differences alone, that I’ve written, and what’s their worth?

I broke it aside this manner as a result of it’s very clear from doing this evaluation that diff-in-diff essays are one of the valued elements of the substack, even going again to day one, and since I randomize the paywall on all my non-Claude Code essays, I even have been capable of get an elasticity measure of the causal impact of difference-in-differences essays even on paid subscribers. Curiously, earlier than Claude Code, the outdated AI writings have been actually not very attention-grabbing or valued by readers. My writings on AI which have been valued have been solely about Claude Code and AI brokers, not AI typically. However let me simply begin on the prime.

Complete Essay Classification

Once more this was completed utilizing Fable 5 and I concentrate on whole subscriptions, not paid versus free. And I used the pre-Claude period simply to make it easy. However there are 9 classes of essays I’ve written on right here numbering astonishingly round 667 whole essays. That’s excluding Claude Code, thoughts you, which provides one other 55. However whole essays, which I assume consists of some stragglers that don’t actually belong wherever, is 806 essays since March 2021 after I began the Substack plus one other 31 essays that don’t have any metadata apparently completely 837 essays.

Here’s what every of those are. Like I stated, for these I gained’t hyperlink because it’s too many to maintain observe of.

1. Diff-in-diff explainers & pedagogy (74 posts, 63.5 subs/publish). The most well-liked style are the diff-in-diff explainers. A lot of the worth presumably of the complete substack is diff-in-diff. Which is what I suspected; I’m not a public mental and don’t write “huge concepts” items. I see myself largely as a tutor and a helper, and diff-in-diff being the preferred quasi-experimental methodology with probably the most progress in econometrics during the last 8 years, it’s not a shock to me both since I’ve intensively invested in studying it and speaking about it. This substack style has actually helped me construct a full curriculum round instructing diff-in-diff too.

2. Different econometrics explainers (45 posts, 25.2 subs/publish). Then there are the extra basic causal inference and econometrics explainers. These are additionally translation essays, however exclude difference-in-differences. It’s issues like my matching-vs-regression “Acts” collection, my Tymon Słoczyński’s OLS weights (”Lies, Rattling Lies and OLS Weights”), matrix completion, LATE and compliers, LaLonde’s legacy, “Which variables do I want to manage for?” Claude stated that these essays “learn like a professor pondering via an idea till it clicks, then handing readers the clicking”. Curiously, he stated that is really the fashion of an essayist. Apparently, the essayist has been traditionally a style that has the texture of “pondering out loud” going again to the very first essayist’s writing. Which I didn’t know — I simply have at all times wanted to put in writing as my mind-set, and it’s at all times a sort of “pondering out loud” versus having already figured it out. My most satisfying essays even have at all times been single drafts in a single sitting the place some downside exists and I work my approach via it. I believe it truthfully could possibly be the ADHD as my mind goes so quick on a regular basis, and I can’t actually focus, however writing forces me to focus. Coding additionally has at all times pressured me to focus, which is partly why Claude Code has been such a double edged sword, however that’s for an additional time. Level is, extra basic explainers on econometrics is a core a part of the Substack’s worth.

3. Analysis commentary (14 posts, 24.3 subs/publish). Then there are outdated issues I don’t do as a lot anymore the place I’m a shopper and producer of analysis relatively than a tutor or trainer of strategies:. That apparently is my outdated now defunct #JHR_Threads explainers about different individuals’s papers revealed within the Journal of Human Assets again after I one of many co-editors there, in addition to my very own papers. I had forgotten about an outdated essay, which Claude says is definitely his favourite, known as Not Waving however Drowning, named after a Stevie Smith poem, about psychological well being struggles amongst graduate college students. There’s additionally my annual Nobel prediction ritual, numerous essays about meta-science worries and empirical crises. Claude stated that this can be a small style with a excessive hit charge. In order that was attention-grabbing to study.

4. Mailbag / recommendation (16 posts, 21.2 subs/publish). Then there was the mailbag which I had some actual momentum with, however simply couldn’t sustain with. I’ve employed a brand new assistant, although, and am onboarding her this week to assist me with my life. And so possibly she will help me get this again going. Anyway, these have been posts with questions answered in public and it included issues like profession recommendation, strategies dilemmas, and many others. It was kind of like an economist’s model of an recommendation column — like a “Pricey Economist” column. Who knew that there was such a style! However there’s. Claude known as this style “Quietly environment friendly: modest quantity, above-average conversion, as a result of the format ensures relevance (somebody actually requested).” That was additionally attention-grabbing to study to be sincere.

5. Workshop & Mixtape enterprise (109 posts, 15.4 subs/publish). Then there are my “business scaffolding” (his phrases) essays, if you wish to name them that. Posts might be a greater phrase than essay for these as these are issues like my Mixtape Periods bulletins, CodeChella, workshop scheduling, ebook revision countdowns, Mixtape Fellows introductions. They’re largely logistics however with my fashion of writing and voice. The common is definitely inflated by one outlier, although — the 2021 CodeChella announcement (630 subs). When you strip it out, then the style is pretty modest at round ~9.7 subs/publish, which is what bulletins ought to do in all probability.

6. Historical past of economics (20 posts, 14.9 subs/publish). Now these have really been a few of my absolute favourite essays if I’m being sincere. It’s the “storyteller style”. Claude famous my Ken Chay essay for example the place I argue that he was the bridge from Rubin to Princeton to Berkeley by way of creating class content material that has change into fairly standardized. But it surely’s additionally my essays about Bob LaLonde like one on his job market paper, the story of artificial management, the Princeton IR part lineage extra usually, the affect of Orley Ashenfelter general. Claude calls these essays as “economics as a human drama with characters, rivalries, and luck.” A whole lot of this materials, really, made it into the brand new ebook too. I nonetheless hope to someday possibly write a extra full “biography of causal inference” (much less so a real “historical past of causal inference” than the type of biographical story that I desire telling) ebook, however for now I’m simply attempting to maintain my head above water.

7. AI experimentation (69 posts, 14.6 subs/publish). Then there have been the pre-Claude Code essays about AI. These are from 2023–2025 and so they functioned as a pure runway into Claude Code. And a few of them have been undoubtedly tremendous bizarre, too. Like my outdated artistic story during which I co-wrote a brief story with ChatGPT within the fashion of one in every of my favourite ebook’s by Lee Smith, Truthful and Tender Women, which is written within the epistolary fashion. It nonetheless was written from the primary individual by that ebook’s most important character Ivy Rowe, however on this one, Ivy Rowe is mainly Susan Athey kind of, and it tells the story of her successful the John Bates Clark award and in the end the Nobel Prize, however in the end being in love with a younger time collection economist, who isn’t actually superb at his job, and the way proud she is of his work during which he fashions the unfold of rabies utilizing vector autoregression. And no, that’s not Guido Imbens, however it was enjoyable to do anyway as I favored how honest the Ivy Rowe character was in her affection for this man’s work when clearly that’s not a really promising analysis agenda, and the way she simply casually mentions that she wins the John Bates Clark award and Nobel Prize, and the way proud he’s of her, however she may be very nonchalant and humble about it. However then there’s different essays in there involving GPT-4 writing and grading its personal econometrics exams, my outdated ElectionGPT prediction stuff, the Economics of AI course diary, my Acemoglu explainers, therapy-and-companionship reflections. Claude known as these “Tinkering in public — half lab pocket book, half marvel.” These have been kind of, in hindsight, a method of essay during which I used to be rehearsing how I’d write Claude Code essays.

8. Podcast & interviews (189 posts, 4.6 subs/publish). These are literally the highest-volume style as a result of that is one thing surrounding the “The Mixtape with Scott episodes” which incorporates the video interviews but in addition the transcripts. It began out as an oral historical past of recent economics, one economist at a time, however this yr it’s been a brand new pivot with Caitlin Myers about Claude Code which I’m a bit of late updating as a result of journey. These mainly have completely no impact on individuals turning into subscribers to the substack, however that additionally isn’t actually the purpose of the substack anyway, not to mention the podcast. It’s only a factor I like doing (in addition to the substack too).

9. Private diary (52 posts, 4.3 subs/publish). Then there’s my essays which might be like a life-log. Me driving to Boston with my dad on my thoughts, the essays about my San Sebastián summers, the primary days of sophistication at Harvard, the Adler-vs-Becker private philosophy collection, my dad’s obituary.

10. Hyperlink roundups (79 posts, 2.9 subs/publish). After which my weekly “Saturday hyperlinks” during which I publish hyperlinks primarily based on a large browser of open tabs. This has completely no impact on making individuals subscribe, however I get pleasure from it.

Share Scott’s Mixtape Substack

Phrases

There have been 1.66 million phrases written throughout 806 posts (excluding the 31 I discussed that don’t have metadata), however Claude says that the “cleaner quantity” is 1.52 million phrases throughout 613 essays with a mean variety of phrases at round 2,500 phrases per essay. The remainder are podcast pages which have gotten round 89,000 phrases, however that’s largely present notes and transcripts, video posts (41,000) and a few advert hoc electronic mail issues. However, 1.5 million phrases is about eight books the dimensions of my ebook, Causal Inference: the Mixtape in line with Claude (undecided how he is aware of that truly) or “two and half Conflict and Peaces”. Right here’s the annual breakdown.

Two issues bounce out. First, my output on right here doubled yearly via 2024. I had half 1,000,000 phrases that yr, which Claude known as “mainly a full-time author’s annual output”. And this yr seems to be on tempo with my largest yr but with 245k phrases in 27 weeks which may annualize to 475,000 phrases if I hold that up. The Claude Code part alone apparently is 152,000 phrases throughout 55 posts which he says “is already a book-length manuscript sitting my archive”. However, I’m skeptical that there’s a coherent ebook in all these. Nonetheless, nobody believes in me greater than 1) my mother, and a couple of) Claude. So after I run this previous my mother, solely then will I contemplate it.

Distinction-in-Variations

Lastly, diff-in-diff. That is the classification of that individual style of essays which above I stated has been a reasonably dominant supply of worth, and a wierd change in the midst of my profession. Essays about nothing extra than simply working my approach via the myriad labyrinth of latest difference-in-differences materials that has come out since 2018, however then working backwards from there too. There are 4 kinds of essays I’ve written about diff-in-differences. They’re right here and I’ll clarify in a second what they’re.

The explainers are probably the most helpful style I’ve ever written, full cease, measured by way of subscriptions and different issues within the metadata that’s an excessive amount of to get into right here. The March 2021 essay Callaway and Sant’Anna explainer introduced in 987 subscriptions which continues to be my single largest publish ever; even greater than Claude Code Half 1 essay which Claude stated introduced in 745. The “Steady therapy DiD” essay was additionally very fashionable (437) as have been the three minimal wage papers essay (275) and the Solar–Abraham essay (216). Ten 2021 DiD posts alone generated 2,713 subs which is 1 / 4 of all the pieces the pre-CC corpus ever earned.

However there was a decay curve contained in the style. Diff-in-diff subs per yr went like this: 2,713 (2021) → 261 → 419 → 809 → 494, even whereas publish quantity rose. The guidelines period (2024–25) produced twice as many posts because the explainer period for a sixth of the subscriptions. This isn’t in line with Claude a high quality judgment. It’s really the identical actual patterns because the Claude Code collection. As soon as the diff-in-diff revolution grew to become widespread data and absolutely absorbed, kind of, my very own writings on the style shifted in direction of essays that existed purely to service the core subscribers. Let me break these down now although.

1. Frontier paper explainers (17 posts, 177 subs/publish). The 2021–22 core was mainly one sizzling paper per publish translated on the actual second when utilized researchers appeared to need it. Like I stated, the primary instance was my first Callaway–Sant’Anna essay which has had 987 subs and is my greatest publish ever. But it surely’s additionally the continual diff-in-diff essay (the primary one I did) which has had 437 subscriptions, the minimal wage wars (275), Solar–Abraham (216). The components, in line with Claude, had a type of “urgency” and “accessibility” really feel to it partly as a result of the frontier of utilized econometrics was shifting very quick, the papers have been type of onerous to learn, they have been mounting up, and I used to be apparently early in offering readable maps. Nothing has had extra affect on the substack than these apparently. I kind of sensed it, however didn’t know that.

2. Instructing collection (21 posts, 36.9 subs/publish). These are multi-part idea arcs written as slow-motion lectures. They embody, for example, my Triple Variations essays (Elements 1–5), mechanisms and ATT, short-gap vs. lengthy variations, inhabitants weights, decomposing TWFE. Whereas the explainer essays have been browsing alongside another person’s particular estimator or paper, the “instructing collection” essays are the place I used to be publicly attempting to construct a pedagogy round instructing these estimators (which in the end funnels into my workshops, my lessons, but in addition the brand new ebook, Causal Inference: the Remix). Apparently, Claude thinks of those essays as resembling “a textbook being drafted in public.” Which curiously is precisely what they grew to become!

3. Observe commentary (14 posts, 32.3 subs/publish). These have been opinions from the workshop ground. They’re in all probability the closest I really ever get to offering robust opinions too. For example, that is the place I wrote about why falsification issues as a lot as occasion research, the right way to visualize an event-study plot, why visible pre-trend checks will be deceiving, what DiD quotation patterns reveal. They have been sometimes quick, opinionated, and kind of primarily based on my experiences, relatively than on what others had written. Claude calls them “the utilized econometrician’s equal of an editor’s column, and the sub-genre the place your authority (relatively than the literature’s novelty) carries the publish.” That was attention-grabbing. And see, not a ton of them, and so they’re positively not as impactful because the explainers.

4. Guidelines / design-stage apply (22 posts, 21.0 subs/publish). The 2024–25 codification mission which I known as “Pedro’s guidelines steps 1 via 4(i)” . I mainly received caught on this guidelines collection as you possibly can see at step 4 the place I ended up writing 4(a) all the best way to 4(i). Typical ADHD. Additionally consists of the “Designing your Diff-in-Diff” collection, implementing the guidelines on county crime information. It’s in the end probably the most posts with the fewest subs per publish of the 4, however Claude stated that his was as a result of by then the diff-in-diff revolution had kind of plateaued and been broadly absorbed. Discover that these are 2024 and 2025. So these have been extra service to the core subscribers, versus chatting with the frontier, so to talk. It’s additionally, not coincidentally, the direct ancestor of “Claude Code 53: Utilized econometrics would require an in depth guidelines”, which stays an upcoming collection I can be doing on utilizing checklists with AI Agentic analysis.

Share

Conclusion

In order that’s it. In case you have been writing a substack for some time, I encourage you to take a look at the brand new Claude integration. It’s fairly illuminating. Mainly Claude Code studying your complete corpus of non-academic public writing after which doing no matter evaluation you need on it.

Claude additionally satisfied me that it was in all probability time to cease numbering the Claude Code collection, which is why I didn’t quantity this one. So going ahead, I’ll simply begin writing with out numbering. And I’ll proceed to maintain the brand new Claude Code essays “free” with out the randomized paywall that I exploit on all the pieces else. Every part finally goes behind the paywall, however the Claude Code ones will proceed to be free instantly primarily based on suggestions I’ve gotten in my travels and in emails. Thanks once more for all of your help!