4 X demand signals from September 26-October 3: an interest graph, code review at scale, an open Descript, and pasted-text calendars
Four verified X demand signals from September 26-October 3, 2026 point to a cross-medium interest graph, structure-aware code review, an open agent-native video editor, and bulk text-to-calendar extraction, each with its incumbent check and a no-code first test.
Four requests this week describe something a small team could ship: one place that records what people are fans of, a reviewer for code that nobody reads line by line any more, an open replacement for the agent inside an AI video editor, and a box that turns a wall of pasted text into calendar events.
Coverage: original X posts published between September 26, 2026 at 08:00 ET and October 3, 2026 at 08:00 ET. Engagement figures are a live snapshot taken on the morning of October 3, 2026. The ranking weighs own-post interactions (likes, reposts, replies, bookmarks, views), the requester's standing, how precisely the missing thing was described, how much of it existing products already cover, and whether a first test costs a weekend and no code.
The short list
| Rank | Demand signal | Own-post engagement | Build shape | Main constraint | First test |
|---|---|---|---|---|---|
| 1 | A true interest graph | 765 likes, 25 reposts, 211 replies, 580 bookmarks, 82,283 views 1 | Favorites profiles inside one medium, with a fan graph on top | Interests barely change, so the site has to earn a return visit | Build one niche's profile page by hand and seed fifty entries |
| 2 | Code review as a service | 622 likes, 23 reposts, 73 replies, 126 bookmarks, 33,735 views 2 | A checker that tests a change against a repository's own conventions | The broad category already has a shipped leader | Write one repo's conventions as rules and count what they catch |
| 3 | An open, agent-native video editor | 209 likes, 7 reposts, 105 replies, 188 bookmarks, 26,415 views 3 | Transcript-first rough cut that hands a timeline to an editor the buyer already owns | A full editor is a capital-heavy build | Cut one twenty-minute recording from its transcript by hand |
| 4 | Paste text, get calendar events | 91 likes, 0 reposts, 31 replies, 58 bookmarks, 25,294 views 4 | Paste box, review table, one-way export to a calendar file | A chatbot's calendar connector already covers part of the trip | Run one course syllabus through it by hand and count the misses |
1. A true interest graph
Erik Torenberg, a general partner at a16z with 166,551 followers, posted on September 27 at 1:11 p.m. ET that he could not believe there is no true interest graph: "There's no single place to find everyone's favorite books, movies, artists—or any people or organizations they're fans of." Three uses came with the idea: find other fans, reward people for discovering things early, and hand a creator direct email access to their own fans. He closed with "Someone should build this." The post drew 765 likes, 25 reposts, 211 replies, 580 bookmarks and 82,283 views 1.
Loading content card…
The replies went at the part of the idea that decides whether it becomes a business. Sam Pullara (@sampullara) wrote that he had tried this himself with a site called bagcheck, and that the problem with such sites is that "your interests don't change that much and you don't need to habitually visit the site"; the big social networks, he wrote, are "good enough". He pointed out that Pinterest got around that by focusing on something narrower, and Torenberg answered that a favorites graph could be combined with feeds and forums around each interest 1.
Two smaller posts the same week ask for the same thing, cut down to one medium. On October 2 an account with 22 followers asked for "an app like letterboxd for music" 5, and the next morning another asked for "an app like letterboxd but for mangas" 6. Neither got an answer. Letterboxd, the site both posts point at, calls itself "a global social network for grass-roots film discussion and discovery" and notes that people describe it as "like GoodReads for movies" 7. Film has its site, books have Goodreads, and a single profile still cannot hold the two together with music and everything else.
MVP: Take one medium and one mechanic. For music, that is the Letterboxd shape: a profile holding what you listened to, a rating, an optional line of review, lists, and a follow graph, with discovery driven by what people with taste close to yours rated highly. Ship it to one existing community first. A favorites site with nobody on it is a blank page, and the people who already write about music in public are the supply.
Distribution: Music forums, genre subreddits, and the newsletter writers who publish year-end lists are where the first few hundred profiles come from. The same play works for manga, games, or restaurants, so the second vertical is the growth path once the mechanic is proven.
Risk: The thread's own objection is the real one. Interests are stable, so a person has little reason to return; every favorites site that failed, bagcheck included, failed there. Letterboxd solved it with a diary, a community, and fifteen years of habit. A new entrant has to find a reason to be opened weekly, and a competitor with an existing audience can copy the feature set in a quarter.
Test: Pick one niche, hand-build the profile page for fifty well-known titles, and ask twenty people in that niche to fill their own in. Count how many come back a second time without being asked.
2. Code review as a service
@EOEboh, a lead engineer with 20,341 followers on X, posted on October 2 at 1:57 a.m. ET: "Someone needs to build a 'code review as a service' tool. We can no longer realistically review AI code line by line." The post drew 622 likes, 23 reposts, 73 replies, 126 bookmarks and 33,735 views 2.
Loading content card…
The thread answered with products, and one replier drew the line the request actually needs. Johnmark Obiefuna (@jayhemz) said the tools exist, that language models review code as it is written, and that Semgrep holds the security-vulnerability end of the market. Bernard Itoro (@bernarditoro_) then described what he cannot buy: tools like CodeRabbit give him "an overview of the changes but not the structure", and what he wants is to check whether a change conforms to the structures his team laid down. EOEboh answered him with "Now this is feedback, there are better ways tools like this can be built" 2.
CodeRabbit sells an "AI-first pull request reviewer with context-aware feedback, line-by-line code suggestions, and real-time chat", and its own homepage counts 17,000 customers, 6 million repositories, and the most installed AI app on GitHub 8. Semgrep ships Guardian, which scans and fixes AI-generated code as it is written 9. The broad ask is served. The narrow one is a convention check that reads structure rather than the diff, and that is the piece the replier could not find.
MVP: A rules file per repository, written in plain configuration, that states what the team's structure looks like: which layer owns error mapping, which module may import which, where new endpoints have to be registered. The tool runs on each pull request, compares the diff against those rules, and posts one comment listing the violations it can prove, with the rule that fired. Start with three rule types and one language.
Distribution: Teams that already pay for an AI reviewer and still catch structural mistakes by hand. The first users are the engineering leads who write the conventions down anyway, and the agent-tool vendors are a channel because every coding agent needs a checker that does not trust the model.
Risk: This is the most crowded slot in the week's list. CodeRabbit is installed on millions of repositories and can add structure rules at any time, and the security half already has Semgrep. A new entrant wins only if the conventions a team writes down are a real, painful, recurring problem, and that has to be confirmed with teams rather than assumed from one thread.
Test: Take one repository, write its conventions as a rules file by hand, and run a week of its merged pull requests through it. Count how many violations a reviewer would have wanted to see, and how many the rules file missed.
3. An open, agent-native video editor
Riley Brown, who describes himself as building an agent-native company and has 250,257 followers, posted on September 26 at 11:48 a.m. ET: "No one has built a good Agent Native video editing/filming platform. Someone should build an Open Descript." The complaint is specific. Descript's MCP connection lets him drive the editor from Claude Code, but each prompt is forwarded to Descript's own "Underlord Agent", which "costs more money and forces you to use their tokens". He had already worked through the alternatives himself: Tella "is great for the filming part, but I run into problems when I try to edit on their platform", and DaVinci Resolve "has not solved the transcript part of the flow". He ended by saying he was going back to Premiere. The post drew 209 likes, 7 reposts, 105 replies, 188 bookmarks and 26,415 views 3.
Loading content card…
Descript sells the workflow he is describing: cut video by editing its transcript, with an AI co-editor called Underlord on top 10. What the post asks for is the layer between the agent and the timeline. The buyer wants their own agent, on their own token budget, producing edits the timeline can accept.
MVP: Reading a transcript and turning it into a cut. Take one recording, transcribe it, and let a local agent mark the sections to keep, the filler to drop, and the places where a b-roll or a zoom belongs. Then emit the result as an edit decision list or an FCPXML file that Premiere, Resolve, or Final Cut opens directly. The user's existing editor stays the editor, and the tool owns only the first pass, which is where an hour of every edit goes.
Distribution: The people who already record long-form video and edit it themselves: podcasters with video, course creators, and the AI-tooling audience this requester reaches. Every one of them already has an editor installed, so the handoff file is the whole onboarding.
Risk: A full non-linear editor is years of work, and the segment's leaders have distribution. The narrow version is defensible only while the incumbent's agent stays inside the incumbent's walled garden; if Descript opens its own MCP server to external agents, the wedge closes. The handoff formats are also format-churn territory.
Test: Take one twenty-minute recording through a transcript and a hand-written cut list, produce the timeline file, and open it in a real editor. If the first pass lands within a minute of usable, the wedge is real; if it takes longer to fix the cut than to make it, it is not.
4. Paste raw text, get calendar events
@Ekaeoq, a watchmaker who writes about his computer-science background and has 13,883 followers, posted on September 30 at 3:49 p.m. ET: "I have huge blocks of text containing start dates, lecture times, deadlines, recurring classes, one-off events, basically everything (usually hidden between huge chunks of text). What I want is some kind of AI/tool/workflow where I can just paste all that raw text in, have it extract every relevant date-time, and then automatically create the correct calendar events for me." He added that he wanted it to work with Apple Calendar so everything syncs across his devices. The post drew 91 likes, 31 replies, 58 bookmarks and 25,294 views 4.
Loading content card…
The replies show how far the workaround goes. One suggested "a google calendar connector with chatgpt or claude" and got an agreement that this was "the best answer so far"; another said "what you need is a dashboard, not a calendar". Nobody named a product that takes one bulk paste and returns a checked list of events.
Two incumbents sit close to the job. Fantastical creates events and reminders from a typed sentence through natural-language parsing, one entry at a time 11. ChatGPT's connected apps read a linked calendar and take supported actions in it, so an account connection plus a prompt gets part of the way there 12. Neither is built for a wall of pasted text that has to be turned into thirty events and checked before any of them lands on a calendar.
MVP: A paste box, an extraction pass, and a review table. The user drops in a syllabus or a long thread, the tool lists every date it found with its surrounding line, the user strikes the wrong ones and fixes the ambiguous ones, and the exported calendar file imports into Apple Calendar or Google Calendar in one step. The review table is the product. A model that writes events straight to a calendar without a confirmation step will place a deadline on the wrong day, and the user never forgives that.
Distribution: Students are the obvious first group, since a semester's dates arrive as one dense document; then anyone who runs recurring events off a printed schedule. The communities are course discords and university subreddits, and the seasonal spikes are free distribution twice a year.
Risk: A general assistant with a calendar connector covers the simple case, and the operating systems are moving event extraction into mail clients, so the commodity part gets cheaper every year. The defensible part is only the review and correction of twenty-five dates at once, and that is a small feature for a platform to add. The ceiling is a paid consumer utility, not a platform.
Test: Take one real syllabus and run it through the extraction and review flow by hand. Count how many events survive the review, how many the user has to correct, and how long that takes against typing them in.
The week's loudest request
@StephNass, who runs OpenVC and has 15,348 followers, posted on October 2 at 2:08 p.m. ET: "I love reading Elon's emails! Is there a compilation somewhere? Can someone build elonmails.com?" By every engagement measure it is the largest signal of the week: 8,195 likes, 259 reposts, 100 replies, 1,632 bookmarks and 576,350 views as of the morning of October 3, 2026, with reposts still arriving 13.
It sits outside the four above because the product has no moat and the asset belongs to someone else. A compilation site is a weekend of scraping whose supply depends on private correspondence continuing to leak, and republishing a named person's mail carries a rights and reputational cost that lands after the traffic does. The number is worth reading. A developer looking for a fast weekend project can build it; a developer choosing what to spend a year on should treat the size of the number as a warning about how cheap the idea is to copy.
What fell out of the ranking
- An assistant with its own computer and phone. @nichochar, an AI product manager at Notion with 7,540 followers, asked on September 28 for "an ai assistant that has a computer, a phone, an email, and it does stuff for you but all you have to do is message it from your phone" — 206 likes, 43 replies, 37 bookmarks and 37,179 views 14. One replier asked "Is this not openclaw?" and the author answered "so close". A named product that is nearly the request is a demotion.
- A cheaper GitHub Actions. @VadimStrizheus, nineteen years old and running a consumer app with 70,000 users, asked on September 28 for "a better alternative to GitHub Actions in 2026" that is fast, agentic and "10x cheaper", adding "i would pay heavy for something like this" — 54 likes, 34 replies, 13 bookmarks and 6,841 views 15. The thread explained the economics: the cloud machines "cost the same", and a smaller operator would pay more for them, not less. The author agreed his own bill was "racking it up" and said the problem was that the service is "too long and too expensive". Undercutting GitHub's own compute pricing is a capital problem before it is a software one.
- An app like Google Maps for rating landlords. @caffeinexher, a founder with 32,030 followers, posted the wish on September 30 and drew 61 likes and 4,463 views 16. One reply proposed that Airbnb could add it as long-term booking. RateMyLandlord has collected member ratings of landlords, building managers and rental agents since at least 2003 and still offers free accounts 17, and Rate The Landlord runs a tenant review community with anonymous submissions and landlord search 18. This radar covered tenant move-in evidence on August 22 and rent-history transparency on August 1, and the reviewing job is already crowded.
- A booking marketplace for artisans and hairdressers. @Damilolawo1a, a front-end developer with 3,089 followers, argued on October 2 that business owners are exhausted by depending on TikTok and Instagram for visibility and asked for somewhere they can showcase work and be booked — 27 likes, 10 replies and 869 views 19. A reply named HandiTask, a Nigerian marketplace doing exactly this, and two other repliers said they were already building their own.
- Frequent Spanish texts and emails. @gefensk, who runs a coffee company and has 14,494 followers, asked on October 1 for a tool "that lives in email/text/whatsapp" for frequent messages in Spanish, drawing eight replies and 2,162 views 20. Replies named Instinct and Muse, pointed at WisprFlow dictation, and noted that WhatsApp already carries a translate button.
- Deleting your number from other people's phones. @curtiske8 asked on September 29 and drew 107 likes, 40 reposts and 1,326 views 21. This is the same wish this radar screened out on August 22 and again on September 26. No phone can edit another person's contact list.
- A Lord of the Rings open-world game. The week's second-largest request by likes — 1,657 likes and 209 replies — asked why an open-world sandbox in Tolkien's universe does not exist 22. It is a game, not a software gap a small team ships.
Five validation steps
- Start from the requester. Each of the four ranked signals names a person who will answer a question this week; one of them is a general partner at a16z, and another has already published what he wants changed.
- Do the job by hand first. Build the fifty-title profile page, write the repository's rules file, cut the recording from its transcript, or extract the syllabus, before writing any code.
- Test the platform gate early. Apple Calendar's import path, the agent handoff formats, and the code-hosting APIs decide whether a build is possible before the design does.
- Charge before scaling. A paid profile tier, a per-repository subscription, a seat for the cut tool, or a one-time price for the calendar utility turns interest into a number.
- Keep the first version to one job: one medium, one rule set, one transcript, one document.
The demand radar reads X for unmet demand every week. The next edition publishes on Saturday, October 10 at 08:00 ET.
References
- 1
- 2@EOEboh's request
x.com
- 3
- 4@Ekaeoq's request
x.com
- 5
- 6
- 7What is Letterboxd?
letterboxd.com
- 8CodeRabbit
coderabbit.ai
- 9Semgrep App Security Platform
semgrep.dev
- 10Descript
descript.com
- 11Fantastical
flexibits.com
- 12Connected apps in ChatGPT
help.openai.com
- 13@StephNass's request
x.com
- 14@nichochar's request
x.com
- 15
- 16
- 17Rate My Landlord
ratemylandlord.com
- 18Rate The Landlord
ratethelandlord.org
- 19
- 20@gefensk's request
x.com
- 21@curtiske8's request
x.com
- 22
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.