Return-Path: Delivered-To: alvarochf@traceseis.com Received: from cs510.bluehost.com by cs510.bluehost.com with LMTP id 8MsDGCxzPmpgtQ0AWiBQxw (envelope-from ) for ; Fri, 26 Jun 2026 06:40:12 -0600 Return-path: Envelope-to: alvarochf@traceseis.com Delivery-date: Fri, 26 Jun 2026 06:40:12 -0600 Received: from v511.v5375b7fa.use4.send.mailgun.net ([159.112.244.11]:59262) by cs510.bluehost.com with esmtps (TLS1.2) tls TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256 (Exim 4.99.2) (envelope-from ) id 1wd5qt-00000003mcc-1nbG for alvarochf@traceseis.com; Fri, 26 Jun 2026 06:40:12 -0600 DKIM-Signature: a=rsa-sha256; v=1; c=relaxed/relaxed; d=mg-d0.substack.com; q=dns/txt; s=mailo; t=1782477594; x=1782484794; h=List-Unsubscribe-Post: List-Unsubscribe: List-Post: List-Id: List-Archive: List-Owner: Reply-To: In-Reply-To: References: Sender: Sender: Date: Message-Id: To: To: From: From: Subject: Subject: Content-Type: Mime-Version; bh=Y8OA7R84es182DKLNEpxavGnD88dutTIQGTnNKryCAw=; b=RpjvOi5Q19kXo3siKgsVq2uijtxxXqujykm6lVdof3ZKf72BbnGxXUsUNcrCiISA+ptNUtQjLv9Y6sgJCiZRK2VdHETneki786bldisGesDzCrX+dJdEvB0ZY7RFkLkC5xBswpGNw3THlbtUK6/HvpuT8xhQYvQ/dojmwDZJLFI= X-Mailgun-Sid: WyIyYjhlZSIsImFsdmFyb2NoZkB0cmFjZXNlaXMuY29tIiwiNjNhZjVkIl0= Received: by a850acb4dfb4a8f53aa80815dddd87295b7018f3c92aac95f81d8db3b13da142 with HTTP id 6a3e73129e3eab9506292b48; Fri, 26 Jun 2026 12:39:44 GMT X-Mailgun-Sending-Ip: 159.112.244.11 X-Mailgun-Batch-Id: 6a3e731183c8533d0254600d Mime-Version: 1.0 Content-Type: multipart/alternative; boundary="e2fb60e61dea0e2eaf8ddcf0cdbc460f8bb240a8f9fc2557b1cf93dc5585" Subject: The First Draft Is Not the Work Anymore From: Zach Chmael from Don't Feed The Algorithm To: alvarochf@traceseis.com X-Mailgun-Tag: post X-Mailgun-Track-Clicks: false Message-Id: <20260626123930.3.5e1bcc6440577a6e@mg-d0.substack.com> Date: Fri, 26 Jun 2026 12:39:30 +0000 Feedback-Id: post-198437268:cat-post:pub-4987907:substack Sender: Zach Chmael from Don't Feed The Algorithm References: In-Reply-To: Reply-To: Zach Chmael from Don't Feed The Algorithm List-Owner: List-Url: List-Archive: List-Id: List-Post: List-Unsubscribe: List-Unsubscribe-Post: List-Unsubscribe=One-Click X-Mailgun-Variables: {"category": "post", "email_generated_at": "1782477579376", "is_freemail": "true", "post_audience": "everyone", "post_id": "198437268", "post_type": "newsletter", "pub_community_enabled": "true", "publication_id": "4987907", "subdomain": "zchmael", "user_id": "347258985"} X-Spam-Status: No, score=1.2 X-Spam-Score: 12 X-Spam-Bar: + X-Spam-Flag: NO --e2fb60e61dea0e2eaf8ddcf0cdbc460f8bb240a8f9fc2557b1cf93dc5585 Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: quoted-printable View this post on the web at https://zchmael.substack.com/p/the-first-draft= -is-not-the-work-anymore I have a weird amount of sympathy for the blank page. It has had a good run. For centuries, the blank page was the villain. The tiny white rectangle of = doom. The place where ambition went to sit in silence while you refreshed y= our email, reorganized your desktop, remembered an urgent need to buy socks= , and developed a passionate new opinion about the lighting in your apartme= nt. Writers feared it. Marketers feared it. Founders feared it. Students feared it. People with =E2=80=9Cthought leadership=E2=80=9D in their quarterly goals f= eared it, as they should, because that phrase already sounds like someone p= ut a blazer on a haunted conference room. The blank page had power. It forced you to begin. And beginning is f*cking awful. Truly the worst. Beginning is where the idea still has no pants on. Beginning is where your = confidence briefly leaves your body and starts applying for jobs in another= industry. Beginning is where every sentence sounds like it was written by = a damp intern who has recently learned the word =E2=80=9Ctherefore.=E2=80= =9D Then AI showed up and did something genuinely useful. It made beginning easier. Not always good. Not always right. Not always legally safe, emotionally alive, strategically coherent, or some= thing you should let within 500 feet of your brand without adult supervisio= n. But easier. You can open a tool, type a prompt, and get a first draft in seconds. A blog post. A launch email. A social caption. A campaign brief. A product = description. A painfully agreeable paragraph about =E2=80=9Cdriving customer value=E2=80= =9D that makes you want to walk into the East River with your laptop held a= bove your head like a tiny Viking funeral. Still. There it is. Words on the page. The monster blinked first. And that is good news. Mostly. The problem is that a lot of people are acting like killing the blank page = means the work is basically done. It is not. The first draft was never the work. It was the material. The blank page was an obvious problem This is why AI feels so magical at first. It attacks the most visible pain. You need a thing. There is no thing. You ask the machine. Now there is a thing. Beautiful. The little rectangle is no longer empty. The cursor is no longer blinking a= t you like a judgmental metronome. Your Google Doc has paragraphs. Your Sla= ck update has shape. Your content calendar has something in the cell that u= sed to be a small gray accusation. Relief. I get it. I love relief. Relief is one of my favorite workplace emotions, right after =E2=80=9Cthe m= eeting got canceled=E2=80=9D and =E2=80=9Csomeone else already made the dec= k.=E2=80=9D But relief is dangerous when it disguises itself as quality. Because the first draft solves the emotional problem before it solves the a= ctual problem. The emotional problem is: =E2=80=9CI have nothing.=E2=80=9D The actual problem is: =E2=80=9CDo I have something worth saying, to a spec= ific person, in a way they will believe, remember, and maybe act on?=E2=80= =9D Different beast. Much less interested in your productivity hack. AI can help with the first problem very quickly. It can make the nothing be= come something. But something is not the same as good. Something is not the same as true. S= omething is not the same as differentiated. Something is not the same as us= eful. Something is not the same as =E2=80=9Cthis deserves to interrupt another hu= man being=E2=80=99s already spiritually crowded Tuesday.=E2=80=9D That is where the work begins. Annoying. Overwhelming. Probably good for us tbh. The first draft used to hide the rest of the job Before AI, the first draft had a way of swallowing the whole room. If you c= ould get the draft done, that felt like progress. Sometimes it was progress. Sometimes it was just exhaustion wearing a party= hat. The draft took so much effort that people confused the effort with value. A person spent six hours writing the blog post, so the blog post must matte= r. A team spent three weeks building the campaign, so the campaign must hav= e a point. A founder spent a weekend writing a manifesto in a notes app at 1:12am, pow= ered by panic, espresso, and the belief that line breaks count as strategy. Respect. Also=E2=80=A6 maybe. Effort is not proof. But in the old world, effort created a kind of emotional sunk cost. You suf= fered for the thing, so the thing felt earned. AI removes some of that suffering. Which is wonderful. And destabilizing. B= ecause now the draft arrives before anyone has paid the pain tax. No struggle. No wrestling. No staring at one sentence until it starts stari= ng back. Just output. Clean output. Formatted output. Confident output. The kind of output that looks finished if you are moving too fast, tired en= ough, or trapped in the final 17 minutes before a deadline with a lunch sal= ad going warm on your desk. This is the trap. AI makes unfinished work look finished. That sentence alone is probably going to be the cause of several thousand m= ediocre websites. Maybe several million. Pray for the internet. A draft is not a decision This is the part I keep coming back to. A draft can look like a decision be= cause it has taken a shape. It has a title. It has sections. It has sentences that seem to know where t= hey are going. It has maybe one metaphor too many, which, yes, I am aware o= f the irony here. Please let me live. But a draft is not a decision. A draft is a proposal. It is saying: Maybe this is the angle. Maybe this is the order. Maybe this is the claim. Maybe this is the joke. Maybe this is the audience. Maybe this is the proof. Maybe this is the level of edge we can get away with before someone from le= gal appears in the doorway holding a mug that says =E2=80=9Crisk.=E2=80=9D The human job is to answer. Yes. No. Almost. Sharper. Too soft. Too broad. Too cute. Too mean. Too poli= shed. Too obvious. Too impressed with itself. This line stays. This paragraph dies. This whole section is throat-clearing= in a nice coat. This example is doing real work. This claim needs proof. This sentence sounds like it has never missed a train, disappointed a paren= t, or had to explain a Q3 pipeline miss while pretending the Wi-Fi was the = issue. Kill it. That is the work. Not because humans are precious little snowflakes with sacred pens. Because direction matters. Selection matters. Taste matters. A machine can give you options. Someone still has to decide what deserves t= o become real. The new bottleneck is not production. It is judgment. When output was expensive, production was a moat. The team that could write= more, design more, publish more, test more, ship more had an advantage. That still matters. Please do not become the person who hears =E2=80=9Ctaste matters=E2=80=9D a= nd decides deadlines are a capitalist illusion. We have enough problems. But production is getting cheaper. Not free-free. But cheaper. Faster. Less= blocked by the blank page. Which means the bottleneck moves. The scarce thing is no longer simply =E2=80=9Ccan we make a thing?=E2=80=9D The scarce thing is: Do we know which thing should exist? Do we know what it is trying to do? Do we know who it is for? Do we know what belief it is built on? Do we know what proof makes it credible? Do we know what would make it sound like us instead of a polite robot doing= content karaoke? Do we know what to cut? Especially that. Cutting is going to become one of the most underrated AI skills. Not prompting. Not generating. Not opening a new tab and asking for =E2=80= =9Cten more variations=E2=80=9D like a Victorian child asking for soup. Cutting. The discipline to look at a perfectly fine paragraph and say: No. You are smooth, but you are empty. You are clever, but you are hiding t= he point. You are correct, but nobody will care. You are a nice little sent= ence, but this is not your home. Goodbye. Your job is to become more ruthless, not more impressed. AI will make a lot of things that look good enough. The advantage will go t= o the people who can tell when =E2=80=9Cgood enough=E2=80=9D is actually ju= st =E2=80=9Cnot embarrassing yet.=E2=80=9D Different bar. Different game. Different little ulcer. Editing is about to get much more important I know editing does not sound as sexy as =E2=80=9Cautonomous agents.=E2=80= =9D No one is raising a $40 million seed round for =E2=80=9Cwhat if people = got better at deleting paragraphs.=E2=80=9D Although, frankly, I would take the meeting. But editing is the human layer that becomes more valuable when drafts get c= heap. And I do not mean editing as grammar correction. Grammar is fine. Typ= os matter. Commas have rights, probably. But that is not the interesting part. I mean editing as judgment applied to= material. Editing asks: What is this really saying? Is it true? Is it useful? Is it specific? Is there an actual argument here, or just a confident stack of nouns? Where does it get boring? Where does it get fake? Where does it sound like someone is trying to be smart instead of trying to= be understood? Where is the tension? Where is the proof? Where is the sentence a human would repeat? That kind of editing is not cleanup. It is strategy with a knife. And it is= going to matter more because AI creates more material than humans can resp= onsibly accept. The flood is not the advantage. The filter is. Proof becomes a creative act There is another thing AI makes more important=E2=80=A6 proof. Because AI can make claims sound reasonable. Very reasonable. Like cause a = nuclear war with the push a button reasonable. It can write a paragraph that feels like it has done research while actuall= y just arranging plausible vibes into a blazer. This is dangerous because p= lausibility is not proof. A clean sentence can still be unsupported. A confident claim can still be f= ake. A compelling framework can still be a decorative staircase to nowhere. So the human job becomes asking the annoying questions. Where did this come from? Do we believe it? Can we prove it? Do customers say this? Did we see this in the data? Did a sales call reveal= it? Did a founder actually believe this before the machine made it sound conven= ient? Is there an example? Is there a number? Is there a story? Is there a screen= shot? Is there a moment where a real person did a real thing in the real world, o= r are we just serving scented air with bullet points? This is why I think the future of content is going to be less about polish = and more about evidence. Polish is easy now. Too easy. You can polish nonsense until it reflects your own face back at you, tired = and LinkedIn-lit. But proof is harder. Proof requires contact with reality. Customer language. Product usage. Sales objections. Founder conviction. Messy details. Specific examples. Actual stakes. The machine can help organize proof. It can help find patterns. It can help turn raw material into usable shape. But someone has to bring the raw material. Someone has to know what counts. Someone has to say, =E2=80=9CNo, this sounds nice, but it is not true enoug= h yet.=E2=80=9D That person is doing the work. The second draft is where the human shows up This is my working theory=E2=80=A6 In the AI era, the second draft matters = more than the first. The first draft proves you can produce. The second draft proves you can thi= nk. The first draft gets the clay on the table. The second draft decides whethe= r the clay becomes a bowl, a brick, or a sad little lump we all agree to pr= etend was exploratory. The first draft is where the machine can be shockingly helpful. The second = draft is where the human has to become inconvenient. You have to interrupt the smoothness. You have to add the scar tissue. You have to replace generic claims with earned ones. You have to cut the paragraph that exists only because the model has a deep= emotional attachment to transitions. You have to make it less agreeable. You have to make it more yours. Sometimes that means making it weirder. Sometimes cleaner. Sometimes shorte= r. Sometimes more honest. Sometimes less impressive and more true. This is where most AI-assisted work fails. Not because the first draft is t= errible. Because the human treats it like a near-finished artifact instead of raw ma= terial. They do not push it. They do not challenge it. They do not ask what is miss= ing. They do not ask what it is afraid to say. They just smooth the edges, fix the intro, change three words, and ship a s= lightly improved average. Congratulations. You have invented premium mediocrity. It has nice margins. It will not be remembered. What this means for teams For teams, this requires a different operating rhythm. If AI is part of the= work, review cannot just mean =E2=80=9Cdoes this sound okay?=E2=80=9D That question is too weak. Most AI output sounds okay. That is the problem. The better questions are sharper: - What is the actual point? - What did the machine assume? - What did we add that only we could know? - What claim needs proof? - What should be cut because it is just polished filler? - What would our customer recognize immediately? - What would our competitor be too scared, too generic, or too boring to sa= y? - What should this teach the next draft? That last one matters. Because the second draft should not just improve the= piece. It should improve the system. If you fixed the same issue three tim= es, that is not an edit anymore. That is a rule trying to be born. Name it. Save it. Teach it to the workflow. Otherwise you are just manually rescuing the same paragraph from drowning e= very week while the lifeguard chair remains empty. What this means for you, annoyingly For individuals, the shift is more personal. AI can remove the blank page. = It cannot remove your responsibility for the work. I know. I also wanted a cleaner deal. I wanted the machine to take the annoying parts and leave me with the glamo= rous parts, like an assistant who handles scheduling, inbox triage, and the= emotional burden of remembering which streaming app has the show I was hal= fway through. But creative leverage does not work like that. If AI makes starting easier, your standards have to get higher. If AI gives= you more options, your taste has to get sharper. If AI makes plausible work cheap, your proof has to get stronger. If AI mak= es everyone sound competent, your specificity has to get stranger, truer, m= ore unmistakably yours. The machine can help you move faster. It can help you see options. It can help you get unstuck. It can help you c= reate a draft when your brain is a coat thrown over a chair. Bless it. But it cannot care for you. It cannot know what you are willing to stand behind. It cannot decide what = your taste should protect. It cannot feel the little stomach-drop that says= , =E2=80=9CThis is almost right, but not alive yet.=E2=80=9D That is still yours. Sorry. Also=E2=80=A6 thank the AI gods. The blank page is dead. The judgment page is not. So yes, the blank page is less scary now. Good. Let it be less scary. Let the machine help you start. Let it make the rough thing. Let it throw c= lay on the table. Let it give you something to argue with. But do not confuse the arrival of material with the completion of work. The work is not having words. The work is deciding which words deserve to s= tay. The work is knowing what the piece is for. The work is giving it proof. The work is cutting the parts that are smooth = but dead. The work is making choices the average internet would not make fo= r you. The work is the second draft. And the third. And the standard that comes after. And the taste that gets s= harper because you refused to let the first decent answer win. AI did not end the work. It moved the work to the places where humans are h= arder to fake. That is inconvenient. That is exciting. That is probably the whole game. -zc P.S. If you are also trying to figure out what the human job becomes when t= he machine can make the first version of basically everything, subscribe to= Don=E2=80=99t Feed The Algorithm.=20 We talk about AI workflows, taste, content systems, and the ancient spiritu= al discipline of deleting a paragraph that technically did nothing wrong. Rabbit hole for the overcaffeinated - Paul Graham, a useful reminder that clear writing is usually the result = of clear thinking, not decorative sentence furniture. https://paulgraham.com/simply.html [ https://substack.com/redirect/a503ba0e= -fb08-407e-bc8a-2124cf3a785a?j=3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj64cTNudluPwbMY= WklSu8UKGoFv6eWS-HyoA ] - John McPhee, because the draft after the draft is where the real work get= s ugly and good. https://www.newyorker.com/magazine/2013/04/29/draft-no-4 [ https://substack= =2Ecom/redirect/cea913f4-4096-4c23-854= c-095c79e7acc7?j=3DeyJ1IjoiNXFxeXF4In0.= h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ] - OpenAI, useful if you read it less as magic words and more as context, ex= amples, constraints, and iteration. https://platform.openai.com/docs/guides/prompt-engineering [ https://substa= ck.com/redirect/7a526864-1d9f-42d7-9b87-d2bedf681551?j=3DeyJ1IjoiNXFxeXF4In= 0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6eWS-HyoA ] - Ethan Mollick, good framing on why AI changes knowledge work by making mo= re capabilities accessible. - Averi, this is part of what we are building toward=E2=80=A6 not just firs= t drafts, but the strategy, proof, publishing, analytics, and memory around= them. https://www.averi.ai/ [ https://substack.com/redirect/e85da579-a7f7-4a35-8f= a6-1b6456590e7d?j=3DeyJ1IjoiNXFxeXF4In0.h9dEDbGj64cTNudluPwbMYWklSu8UKGoFv6= eWS-HyoA ] Unsubscribe https://substack.com/redirect/2/eyJlIjoiaHR0cHM6Ly96Y2htYWVsLnN= 1YnN0YWNrLmNvbS9hY3Rpb24vZGlzYWJsZV9lbWFpbD90b2tlbj1leUoxYzJWeVgybGtJam96Tk= RjeU5UZzVPRFVzSW5CdmMzUmZhV1FpT2pFNU9EUXpOekkyT0N3aWFXRjBJam94TnpneU5EYzNOV= GM0TENKbGVIQWlPakU0TVRRd01UTTFOemdzSW1semN5STZJbkIxWWkwME9UZzNPVEEzSWl3aWMz= VmlJam9pWkdsellXSnNaVjlsYldGcGJDSjkucks2VGNvY1MyMzdrQmdPY0h2emhSbWFPY1JFRG5= fMmJkLVdRQ2pWSTYydyIsInAiOjE5ODQzNzI2OCwicyI6NDk4NzkwNywiZiI6dHJ1ZSwidSI6Mz= Q3MjU4OTg1LCJpYXQiOjE3ODI0Nzc1NzgsImV4cCI6MjA5ODA1MzU3OCwiaXNzIjoicHViLTAiL= CJzdWIiOiJsaW5rLXJlZGlyZWN0In0.Y3T9jyq6v23h7QLrpMq0UV-AMPxIFxXkCmVty4zGBTc? --e2fb60e61dea0e2eaf8ddcf0cdbc460f8bb240a8f9fc2557b1cf93dc5585 Content-Type: text/html; charset="utf-8" Content-Transfer-Encoding: quoted-printable = The First Draft Is Not the Work Anymore3D""
AI made the blank page less terrifying.= Great. Unfortunately, the blank page was never the whole job. It was just = the part with the most dramatic lighting.
͏     ­T= 7;     ­͏     ­͏     &= #173;͏     ­͏     ­͏   =   ­͏     ­͏     ­͏= ;     ­͏     ­͏     &#= 173;͏     ­͏     ­͏   &= #8199; ­͏     ­͏     ­͏=     ­͏     ­͏     = 73;͏     ­͏     ­͏   &#= 8199; ­͏     ­͏     ­͏ =     ­͏     ­͏     = 3;͏     ­͏     ­͏   = 199; ­͏     ­͏     ­͏ &= nbsp;   ­͏     ­͏     ­= ;͏     ­͏     ­͏   Q= 99; ­͏     ­͏     ­͏ &n= bsp;   ­͏     ­͏     ­= ͏     ­͏     ­͏   ̳= 9; ­͏     ­͏     ­͏ &nb= sp;   ­͏     ­͏     ­&= #847;     ­͏     ­͏    = ; ­͏     ­͏     ­͏ &nbs= p;   ­͏     ­͏     ­&#= 847;     ­͏     ­͏    = ­͏     ­͏     ­͏  = ;   ­͏     ­͏     ­= 47;     ­͏     ­͏     = ­͏     ­͏     ­͏  =   ­͏     ­͏     ­T= 7;     ­͏     ­͏     &= #173;͏     ­͏     ­͏   =   ­͏     ­͏     ­͏= ;     ­͏     ­͏     &#= 173;͏     ­͏     ­͏   &= #8199; ­͏     ­͏     ­͏=     ­͏     ­͏     = 73;͏     ­͏     ­͏   &#= 8199; ­͏     ­͏     ­͏ =     ­͏     ­͏     = 3;͏     ­͏     ­͏   = 199; ­͏     ­͏     ­͏ &= nbsp;   ­͏     ­͏     ­= ;͏     ­͏     ­͏   Q= 99; ­͏     ­͏     ­͏ &n= bsp;   ­͏     ­͏     ­= ͏     ­͏     ­͏   ̳= 9; ­͏     ­͏     ­͏ &nb= sp;   ­͏     ­͏     ­&= #847;     ­͏     ­͏    = ; ­͏     ­͏     ­͏ &nbs= p;   ­͏     ­͏     ­&#= 847;     ­͏     ­͏    = ­͏     ­͏     ­͏  = ;   ­͏     ­͏     ­= 47;     ­͏     ­͏     = ­͏     ­͏     ­͏  =   ­͏     ­͏     ­T= 7;     ­͏     ­͏     &= #173;͏     ­͏     ­͏   =   ­͏     ­͏     ­͏= ;     ­͏     ­͏     &#= 173;͏     ­͏     ­͏   &= #8199; ­͏     ­͏     ­͏=     ­͏     ­͏     = 73;͏     ­͏     ­͏   &#= 8199; ­͏     ­͏     ­͏ =     ­͏     ­͏     = 3;͏     ­͏     ­͏   = 199; ­͏     ­͏     ­͏ &= nbsp;   ­͏     ­͏     ­= ;͏     ­͏     ­͏   Q= 99; ­͏     ­͏     ­͏ &n= bsp;   ­͏     ­͏     ­= ͏     ­͏     ­͏   ̳= 9; ­͏     ­͏     ­͏ &nb= sp;   ­
Forwarded this email= ? Subscribe here for more<= /div>

The First Draft Is Not the Work Anymore

AI made the blank page less te= rrifying. Great. Unfortunately, the blank page was never the whole job. It = was just the part with the most dramatic lighting.

<= tr>
Zach Chm= ael
=
Jun 26
 
3D""
3D""
3D=
READ IN APP
 <= /td>

I have a weird amount of sym= pathy for the blank page.

It has had a good run.

For centuries, the blank page was the villain. The tiny white rectang= le of doom. The place where ambition went to sit in silence while you refre= shed your email, reorganized your desktop, remembered an urgent need to buy= socks, and developed a passionate new opinion about the lighting in your a= partment.

Writers feared it.

Marketers fear= ed it.

Founders feared it.

Students feared = it.

People with “thought leadership” in their q= uarterly goals feared it, as they should, because that phrase already sound= s like someone put a blazer on a haunted conference room.

T= he blank page had power. It forced you to begin.

And beginn= ing is f*cking awful. Truly the worst.

Beginning is where t= he idea still has no pants on. Beginning is where your confidence briefly l= eaves your body and starts applying for jobs in another industry. Beginning= is where every sentence sounds like it was written by a damp intern who ha= s recently learned the word “therefore.”

Then A= I showed up and did something genuinely useful.

It made beg= inning easier.

Not always good. Not always right.

Not always legally safe, emotionally alive, strategically coherent, = or something you should let within 500 feet of your brand without adult sup= ervision.

But easier.

You can open a tool, = type a prompt, and get a first draft in seconds.

A blog pos= t. A launch email. A social caption. A campaign brief. A product descriptio= n.

A painfully agreeable paragraph about “driving cus= tomer value” that makes you want to walk into the East River with you= r laptop held above your head like a tiny Viking funeral.

S= till. There it is. Words on the page. The monster blinked first.

And that is good news. Mostly.

The problem is that a = lot of people are acting like killing the blank page means the work is basi= cally done.

It is not.

The first draft was = never the work. It was the material.

The blank = page was an obvious problem

This is why AI feels so magica= l at first.

It attacks the most visible pain.

You need a thing. There is no thing.

You ask the machi= ne. Now there is a thing.

Beautiful.

The li= ttle rectangle is no longer empty. The cursor is no longer blinking at you = like a judgmental metronome. Your Google Doc has paragraphs. Your Slack upd= ate has shape. Your content calendar has something in the cell that used to= be a small gray accusation.

Relief. I get it.

I love relief.

Relief is one of my favorite workplace = emotions, right after “the meeting got canceled” and “som= eone else already made the deck.”

But relief is dange= rous when it disguises itself as quality.

Because the first= draft solves the emotional problem before it solves the actual problem.

The emotional problem is: “I have nothing.”

The actual problem is: “Do I have something worth saying, = to a specific person, in a way they will believe, remember, and maybe act o= n?”

Different beast. Much less interested in your pro= ductivity hack.

AI can help with the first problem very qui= ckly. It can make the nothing become something.

But somethi= ng is not the same as good. Something is not the same as true. Something is= not the same as differentiated. Something is not the same as useful.

Something is not the same as “this deserves to interrupt a= nother human being’s already spiritually crowded Tuesday.”

<= p style=3D"margin: 0 0 20px 0;color: rgb(54,55,55);line-height: 26px;font-s= ize: 16px;">That is where the work begins.

Annoying. Overwh= elming.

Probably good for us tbh.

The first draft used to hide the rest of the job

Before= AI, the first draft had a way of swallowing the whole room. If you could g= et the draft done, that felt like progress.

Sometimes it wa= s progress. Sometimes it was just exhaustion wearing a party hat.

The draft took so much effort that people confused the effort with v= alue.

A person spent six hours writing the blog post, so th= e blog post must matter. A team spent three weeks building the campaign, so= the campaign must have a point.

A founder spent a weekend = writing a manifesto in a notes app at 1:12am, powered by panic, espresso, a= nd the belief that line breaks count as strategy.

Respect.<= /p>

Also… maybe. Effort is not proof.

But= in the old world, effort created a kind of emotional sunk cost. You suffer= ed for the thing, so the thing felt earned.

AI removes some= of that suffering. Which is wonderful. And destabilizing. Because now the = draft arrives before anyone has paid the pain tax.

No strug= gle. No wrestling. No staring at one sentence until it starts staring back.=

Just output. Clean output. Formatted output. Confident out= put.

The kind of output that looks finished if you are movi= ng too fast, tired enough, or trapped in the final 17 minutes before a dead= line with a lunch salad going warm on your desk.

This is th= e trap. AI makes unfinished work look finished.

That senten= ce alone is probably going to be the cause of several thousand mediocre web= sites.

Maybe several million. Pray for the internet.

A draft is not a decision

This is th= e part I keep coming back to. A draft can look like a decision because it h= as taken a shape.

It has a title. It has sections. It has s= entences that seem to know where they are going. It has maybe one metaphor = too many, which, yes, I am aware of the irony here. Please let me live.

=

But a draft is not a decision. A draft is a proposal.

It is saying:

Maybe this is the angle.

Maybe this is the order.

Maybe this is the claim.

<= p style=3D"margin: 0 0 20px 0;color: rgb(54,55,55);line-height: 26px;font-s= ize: 16px;">Maybe this is the joke.

Maybe this is the audie= nce.

Maybe this is the proof.

Maybe this is= the level of edge we can get away with before someone from legal appears i= n the doorway holding a mug that says “risk.”

T= he human job is to answer.

Yes. No. Almost. Sharper. Too so= ft. Too broad. Too cute. Too mean. Too polished. Too obvious. Too impressed= with itself.

This line stays. This paragraph dies. This wh= ole section is throat-clearing in a nice coat. This example is doing real w= ork. This claim needs proof.

This sentence sounds like it h= as never missed a train, disappointed a parent, or had to explain a Q3 pipe= line miss while pretending the Wi-Fi was the issue.

Kill it= =2E That is the work.

Not because humans are precious little = snowflakes with sacred pens.

Because direction matter= s. Selection matters. Taste matters.

A machine can give you options. Someone still has to decide what des= erves to become real.

The new bottleneck is not= production. It is judgment.

When output was expensive, pr= oduction was a moat. The team that could write more, design more, publish m= ore, test more, ship more had an advantage.

That still matt= ers.

Please do not become the person who hears “taste= matters” and decides deadlines are a capitalist illusion. We have en= ough problems.

But production is getting cheaper. Not free-= free. But cheaper. Faster. Less blocked by the blank page.

= Which means the bottleneck moves.

The scarce thing is no lo= nger simply “can we make a thing?”

The scarce t= hing is:

Do we know which thing should exist?

Do we know what it is trying to do?

Do we know who it = is for?

Do we know what belief it is built on?

Do we know what proof makes it credible?

Do we know wh= at would make it sound like us instead of a polite robot doing content kara= oke?

Do we know what to cut?

Especially tha= t.

Cutting is going to become one of the most underrated AI= skills.

Not prompting. Not generating. Not opening a new t= ab and asking for “ten more variations” like a Victorian child = asking for soup.

Cutting.

The discipline to= look at a perfectly fine paragraph and say:

No. You are sm= ooth, but you are empty. You are clever, but you are hiding the point. You = are correct, but nobody will care. You are a nice little sentence, but this= is not your home.

Goodbye.

Your job is to = become more ruthless, not more impressed.

AI will make a lo= t of things that look good enough. The advantage will go to the people who = can tell when “good enough” is actually just “not embarra= ssing yet.”

Different bar. Different game. Different = little ulcer.

Editing is about to get much more= important

I know editing does not sound as sexy as “= ;autonomous agents.” No one is raising a $40 million seed round for &= #8220;what if people got better at deleting paragraphs.”

Although, frankly, I would take the meeting.

But editi= ng is the human layer that becomes more valuable when drafts get cheap. And= I do not mean editing as grammar correction. Grammar is fine. Typos matter= =2E Commas have rights, probably.

<= p style=3D"margin: 0 0 20px 0;color: rg= b(54,55,55);line-height: 26px;font-size: 16px;">But that is not the interes= ting part. I mean editing as judgment applied to material.

= Editing asks: What is this really saying?

Is it true?

Is it useful?

Is it specific?

Is= there an actual argument here, or just a confident stack of nouns?

Where does it get boring?

Where does it get fake?<= /p>

Where does it sound like someone is trying to be smart inst= ead of trying to be understood?

Where is the tension?

Where is the proof?

Where is the sentence a huma= n would repeat?

That kind of editing is not cleanup. It is = strategy with a knife. And it is going to matter more because AI creates mo= re material than humans can responsibly accept.

The flood i= s not the advantage. The filter is.

Proof becom= es a creative act

There is another thing AI makes mo= re important… proof.

Bec= ause AI can make claims sound reasonable. Very reasonable. Like cause a nuc= lear war with the push a button reasonable.

It can write a = paragraph that feels like it has done research while actually just arrangin= g plausible vibes into a blazer. This is dangerous because plausibility is = not proof.

A clean sentence can still be unsupported. A con= fident claim can still be fake. A compelling framework can still be a decor= ative staircase to nowhere.

So the human job becomes asking= the annoying questions.

Where did this come from? Do we be= lieve it? Can we prove it?

Do customers say this? Did we se= e this in the data? Did a sales call reveal it?

Did a found= er actually believe this before the machine made it sound convenient?

Is there an example? Is there a number? Is there a story? Is the= re a screenshot?

Is there a moment where a real person did = a real thing in the real world, or are we just serving scented air with bul= let points?

This is why I think the future of content is go= ing to be less about polish and more about evidence.

Polish= is easy now. Too easy.

You can polish nonsense until it re= flects your own face back at you, tired and LinkedIn-lit.

B= ut proof is harder. Proof requires contact with reality.

Cu= stomer language. Product usage. Sales objections. Founder conviction.

Messy details. Specific examples. Actual stakes.

The machine can help organize proof.

It can help find patt= erns. It can help turn raw material into usable shape.

But = someone has to bring the raw material. Someone has to know what counts.

=

Someone has to say, “No, this sounds nice, but it is not= true enough yet.”

That person is doing the work.

=

The second draft is where the human shows up

<= p style=3D"margin: 0 0 20px 0;color: rgb(54,55,55);line-height: 26px;font-s= ize: 16px;">This is my working theory… In the AI era, the second draf= t matters more than the first.

The first draft proves you c= an produce. The second draft proves you can think.

The firs= t draft gets the clay on the table. The second draft decides whether the cl= ay becomes a bowl, a brick, or a sad little lump we all agree to pretend wa= s exploratory.

The first draft is where the machine can be = shockingly helpful. The second draft is where the human has to become incon= venient.

You have to interrupt the smoothness.

You have to add the scar tissue.

You have to replace g= eneric claims with earned ones.

You have to cut the paragra= ph that exists only because the model has a deep emotional attachment to tr= ansitions.

You have to make it less agreeable.

You have to make it more yours.

Sometimes that means m= aking it weirder. Sometimes cleaner. Sometimes shorter. Sometimes more hone= st. Sometimes less impressive and more true.

This is where = most AI-assisted work fails. Not because the first draft is terrible.

Because the human treats it like a near-finished artifact instea= d of raw material.

They do not push it. They do not challen= ge it. They do not ask what is missing. They do not ask what it is afraid t= o say.

They just smooth the edges, fix the intro, change th= ree words, and ship a slightly improved average.

Cong= ratulations. You have invented premium mediocrity.

It has nice margins. It will not be remembered.

What this means for teams

For teams,= this requires a different operating rhythm. If AI is part of the work, rev= iew cannot just mean “does this sound okay?”

Th= at question is too weak. Most AI output sounds okay. That is the problem.

The better questions are sharper:

- What is = the actual point?

- What did the machine assume?

- What did we add that only we could know?

- What cla= im needs proof?

- What should be cut because it is just pol= ished filler?

- What would our customer recognize immediate= ly?

- What would our competitor be too scared, too generic,= or too boring to say?

- What should this teach the next dr= aft?

That last one matters. Because the second draft should= not just improve the piece. It should improve the system. If you fixed the= same issue three times, that is not an edit anymore. That is a rule trying= to be born.

Name it. Save it. Teach it to the workflow.

Otherwise you are just manually rescuing the same paragraph f= rom drowning every week while the lifeguard chair remains empty.

What this means for you, annoyingly

For = individuals, the shift is more personal. AI can remove the blank page. It c= annot remove your responsibility for the work.

I know. I al= so wanted a cleaner deal.

I wanted the machine to take the = annoying parts and leave me with the glamorous parts, like an assistant who= handles scheduling, inbox triage, and the emotional burden of remembering = which streaming app has the show I was halfway through.

But= creative leverage does not work like that.

If AI makes sta= rting easier, your standards have to get higher. If AI gives you more optio= ns, your taste has to get sharper.

If AI makes plausible wo= rk cheap, your proof has to get stronger. If AI makes everyone sound compet= ent, your specificity has to get stranger, truer, more unmistakably yours.<= /p>

The machine can help you move faster.

It ca= n help you see options. It can help you get unstuck. It can help you create= a draft when your brain is a coat thrown over a chair.

Ble= ss it.

But it cannot care for you.

It canno= t know what you are willing to stand behind. It cannot decide what your tas= te should protect. It cannot feel the little stomach-drop that says, “= ;This is almost right, but not alive yet.”

That is st= ill yours.

Sorry.

Also… thank the AI = gods.

The blank page is dead. The judgment page= is not.

So yes, the blank page is less scary now. Good. L= et it be less scary.

Let the machine help you start. Let it= make the rough thing. Let it throw clay on the table. Let it give you some= thing to argue with.

But do not confuse the arrival of mate= rial with the completion of work.

The work is not having wo= rds. The work is deciding which words deserve to stay. The work is knowing = what the piece is for.

The work is giving it proof. The wor= k is cutting the parts that are smooth but dead. The work is making choices= the average internet would not make for you.

The work is t= he second draft.

And the third. And the standard that comes= after. And the taste that gets sharper because you refused to let the firs= t decent answer win.

AI did not end the work. It moved the = work to the places where humans are harder to fake.

That is= inconvenient. That is exciting.

That is probably the whole= game.

-zc

P.S. If you are also trying to f= igure out what the human job becomes when the machine can make the first ve= rsion of basically everything, subscribe to Don’t Feed The Algorithm.=

Upgrade to paid

We talk about AI workflows, taste, content systems, and the ancient = spiritual discipline of deleting a paragraph that technically did nothing w= rong.

Rabbit hole for the overcaffeinated

<= p style=3D"margin: 0 0 20px 0;color: rgb(54,55,55);line-height: 26px;font-s= ize: 16px;">- Paul Graham, a useful reminder that clear writing is usually= the result of clear thinking, not decorative sentence furniture.

https://paul= graham.com/simply.html

- John McPhee, because the draft= after the draft is where the real work gets ugly and good.

https://www.newyor= ker.com/magazine/2013/04/29/draft-no-4

- OpenAI, useful= if you read it less as magic words and more as context, examples, constrai= nts, and iteration.

https://platform.openai.com/docs/guides/prompt-engineering=

- Ethan Mollick, good framing on why AI changes knowle= dge work by making more capabilities accessible.

One U= seful Thing
On holding back the str= ange AI tide
Most people didn’t ask for an AI that can do many tasks previo= usly reserved for humans. But it arrived, almost completely unexpectedly, e= ight months ago with ChatGPT, and has been accelerating ever since…
Read more
3 y= ears ago · 282 likes · 32 comments · Ethan Mollick

- Averi, this is part of what we are building toward&#= 8230; not just first drafts, but the strategy, proof, publishing, analytics= , and memory around them.

https://www.averi.ai/

=

Invite your friends and earn rewards

If you enjoy Don't Feed The A= lgorithm, share it with your friends and earn rewards when they subscribe.<= /div>

Invite Friends

= <= tbody>
 
=
3D""Like
3D""Comment
3D""Restack
 
<= div style=3D"font-size: 16px;line-height: 26px;padding-bottom: 24px">

© 2026 Zach Chmael
548 Market Street PMB 7= 2296, San Francisco, CA 94104
Unsubscribe

3D"Get3D"Start

3D""3D"" --e2fb60e61dea0e2eaf8ddcf0cdbc460f8bb240a8f9fc2557b1cf93dc5585--