3 X demand signals from September 19-26: agent trade replay, reusable business verification, and an open-loops digest (2026)

Three requests this week point at gaps a small team could take on: a fund partner who wants to watch his trading agent think, a builder who is tired of proving the same company documents to every platform separately, and an engineer who wants one list of everything his conversations left open. Two of the three come with something a ranking cannot buy — one requester offered to user-test a build inside the thread, another wrote that he would pay.

Coverage: original X posts published between September 19, 2026 at 08:00 ET and September 26, 2026 at 08:00 ET. Engagement figures are a live snapshot taken on the morning of September 26, 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	Live replay of a trading agent	144 likes, 6 reposts, 46 replies, 85 bookmarks, 27,080 views 1	Read-only timeline that replays each fill against the market state the agent saw	Several teams in the same thread say they are already building it	Replay one public agent's last 200 trades and record a three-minute walkthrough
2	Reusable business verification	41 likes, 2 reposts, 9 replies, 2 bookmarks, 2,614 views 2	Document packet with expiry dates and a share link per counterparty	Whoever checks the credential has to accept one it did not issue	Assemble one agency's packet by hand for its next three vendor onboardings
3	A digest of everything left open	19 likes, 2 reposts, 2 replies, 2 bookmarks, 1,069 views 3	Desktop digest built on official APIs	WhatsApp and Android SMS are closed to outside readers	Run the digest by hand for one user for five days from exported data

1. Live replay of a trading agent

Kyle Samani, whose profile lists him as a Multicoin Capital co-founder now building a new company on Solana, posted on September 24 at 11:09 p.m. ET asking whether anyone has built a good real-time visualization of an agent trading: spot trades, perp trades, a timeline, position sizes, P&L, in "some sort of beautiful moving system." The post drew 144 likes, six reposts, 46 replies, 85 bookmarks and 27,080 views 1.

Loading content card…

The replies divide cleanly, and the division is the useful part. One answered "yes" and offered a user test; another named @tmachinroBot as already doing exactly that; a third said it could be built quickly on SpacetimeDB; two more said they were working on it now, one of them for a specific chain. On the other side sat the people describing their current stack: an agent's logs pushed into Grafana, a home-made dashboard, a 3D model. One reply put the state of the market plainly: "surprised this doesn't already exist tbh, feels like an obvious build for anyone running agent trading at scale, would be genuinely useful not just cool to look at" 1.

The reply that should decide what gets built came from a reader who has tried to audit an agent after the fact: "would be especially useful if every trade preserved the market state the agent actually saw. otherwise replaying a bad entry later means judging it against data the agent never had." A second asked for the decision tree behind each trade. The gap in the market is an audit trail with a picture on top; the price chart already exists in a dozen tools. To be clear about what a reader would be entering: at least four groups answered this post with work in progress, so this is a contested slot.

MVP: Your agent logs each decision — the data it had, the order it placed, the size, and a one-line reason. The app pulls fills from the exchange or wallet API, aligns them with those logs, and renders a scrubbable timeline: position size over time, P&L, and the reason behind every entry and exit. Read-only, desktop web, self-hostable, one venue and one agent framework at the start. The market-state snapshot is what makes the replay honest, and it is the part no dashboard does today.

Distribution: Funds and prop desks running agents are the buyers, and the thread is a warm list of roughly a dozen of them, including the account that offered to test. The agent-framework and model-router crowd supplies the small accounts. Price per seat per month.

Risk: The chart layer is commodity work, and the traders' own tooling already covers market data. The scarce asset is the captured state, which means the product improves the earlier it sits inside the agent's stack and gets weaker the longer it waits. If the agent frameworks add their own visualizer, a standalone tool loses its reason to exist.

Test: Take one public agent wallet, replay its last 200 trades against market data, and send Samani a three-minute screen recording of the result. Build the capture hook only if he or one of his repliers asks for access.

2. Reusable business verification

@SimkinStepan posted on September 25 at 6:06 p.m. ET: "someone needs to build zk plaid for reusable business credentials so we can forget the pains of KYB." The post drew 41 likes, two reposts, nine replies, two bookmarks and 2,614 views from an account with 12,216 followers 2.

Loading content card…

Nobody in the thread offered a product. The replies treated the problem as an infrastructure fight that has not started: one said that if the Treasury or Congress ever bless a reusable credential, "there will be a dozen in about 5 minutes"; another said zk technology is the way and that he is "tired of KYC"; one guessed that Metaplex is building it; one replied "launched it" about somebody else's effort. The thread describes a market with demand, no shipped leader, and several people watching the same door 2.

Business verification is sold today to the party asking the questions, one check at a time. Persona packages KYB as a program you run on your own customers 4, and Middesk's API verifies a business on demand for whichever platform calls it 5. Both work for the checker. One product has taken the opposite side: Trustform sells a company-side "business identity wallet" that holds ownership, directors, governance roles and corporate documents, and lets a company share verifiable proofs with banks, investors and platforms 6. That is the request almost word for word. Trustform's page currently routes visitors to early access rather than a signup, which leaves the category declared but unclaimed.

MVP: Skip the zero-knowledge layer. Sell to the checked side first: one place that holds a small vendor's incorporation documents, ownership and UBO list, director identities, proof of address, bank letter, insurance certificate and tax forms, each with an expiry date and an audit log, plus a share link per counterparty and a PDF or JSON export that matches what most platforms ask for. Charge per company per month; the first fix is that the packet never has to be rebuilt.

Distribution: Start where onboarding pain is repetitive — agencies, marketplace sellers, crypto OTC desks, companies opening bank accounts in several countries. The accountants and company-formation agents who assemble these documents today for a fee are the channel, since they already hold the customer relationship and would rather not repeat the assembly.

Risk: A credential hub is worth nothing while checkers insist on their own vendor's result, and some checks stay non-delegable by regulation. The zero-knowledge version of the request is a longer fight than the paperwork version. Whoever ships the checked-side wallet first sets the price, and the category has at least one declared entrant.

Test: Assemble the packet by hand for one agency's next three vendor onboardings, give the reviewer a share link, and count how much of the document list disappears.

3. A digest of everything left open

kanav (@kanavtwt), a 22-year-old engineer with 20,076 followers, posted on September 24 at 5:41 a.m. ET: "can someone build an AI app that has access to ALL my apps and it creates a todo based on the open loops for me? i would pay for this." The post drew 19 likes, two reposts, two replies, two bookmarks and 1,069 views 3.

Loading content card…

The thread produced one suggestion, and the answer to it defines the product. A replier pointed at Gemini, which reaches Google's own apps. kanav answered: "not enough, I'm talking whatsapp, telegram, slack, heck even sms." Take that list at face value and the buildable subset is easy to draw.

Google's Workspace app inside Gemini reads Gmail, Docs, Drive, Calendar, Tasks and Keep 7. Telegram publishes both a Bot API and TDLib, the library third-party developers use to build their own clients 8. WhatsApp's developer platform is a business messaging product, aimed at companies talking to their customers 9. Android grants the SMS permission group only for permitted core uses and requires a declaration form before an app can even ask 10. Three of the four sources he named are workable, one is closed, and the honest pitch covers the three.

MVP: A desktop app that connects Gmail, Slack and Telegram through their official interfaces, reads for commitments and unanswered questions — "I'll send it Friday", "let me check with the team", a question nobody answered — and writes one morning digest with a link back to every thread. Read-only. The human sends the replies. The version that writes on your behalf is a later product with a lawyer attached.

Distribution: Knowledge workers who already pay for a task app and still drop threads, reached through the productivity and developer-tool newsletters. The requester is the first user; the other people who answered him are the second.

Risk: The ask is "ALL my apps" and the deliverable is three, with each new source a permission negotiation that can end in a no. Message data is the most sensitive thing a person can hand an app, so the trust bar is high and a leak is fatal. The OS vendors are the real ceiling: the day Apple or Google ships this inside the system assistant, a standalone tool loses its edge.

Test: For one heavy message user, run the digest by hand for five days from exported data and a manual read. Count the open loops that got closed and would otherwise have been dropped, then ask what that service is worth per month.

A borderline signal

Two of the three ranked entries above are contested, and this small one is the cleanest gap of the week. Alberto Fortin (@a7fort, 616 followers) asked on September 25 at 8:54 a.m. ET for a tool that keeps config.toml and settings.json in sync so Claude Code and Codex share one set of guardrails 11. It drew one like, three replies and 115 views, which is far below the rest of this list. It stays because all three repliers described the same failure and none named a tool: one generates both files from a single AGENTS.md, one said he had found nothing clean and would write a pre-commit script himself, and one wrote the rule out — "two config files that must stay identical will drift within a week unless one is generated."

The pieces are real and separate. Claude Code reads its policy from ~/.claude/settings.json, a project's .claude/settings.json and a machine-local override 12. Codex reads ~/.codex/config.toml with project-level .codex/config.toml files layered on top 13. AGENTS.md is the shared instruction file that most agent tools read, and it carries project instructions rather than each tool's own configuration keys 14. A weekend command-line tool that renders every agent's config from one source, with a diff check in CI, is buildable. Treat it as a small product for a small market, and watch for a second request before committing.

What fell out of the ranking

A decoy agent for spam calls. The week's loudest post came from @APompliano, an investor and writer with 241,301 followers, asking for an AI agent that answers spam calls and keeps the caller talking until the channel stops paying 15. It drew 11,681 likes, 478 reposts and 626 replies. RoboKiller has answered unwanted calls with recorded bots for years and sells the feature to consumers 16. The replies raised the two problems that decide the outcome: many legitimate business calls would be caught in the same net, and the spam operation is already a bot that costs almost nothing to run.
A trusted risk rating for DeFi protocols. @samconnerone, whose account leads developer community work at Circle and Arc, posted on September 26 at 7:37 a.m. ET asking for a rating covering security, liquidity, decentralization, audits and governance 17, and drew 88 likes, 34 reposts, 44 replies and 1,541 views. Replies inside the first half hour pointed at shipped and in-flight work in the same ecosystem: one said Kairo Market "already built this and launched it on arc yesterday", a reply from the @arclens_app side said the feature fits that product and is coming in weeks, and one asked for the feature inside DeFiLlama. The raw material is public already — DeFiLlama tracks TVL, fees and revenue across 7,000-plus protocols 18 and CertiK sells security scores through Skynet 19. The missing ingredient is trust, and the ecosystem that asked for it is supplying its own answers.
Deleting your number from other people's phones. Three posts asked for it in one week, on September 21 20, September 24 21 and September 25 22. No phone can edit the contact list on another person's phone, and the job underneath — hand out a line you can cut off later — is already a product: a second number is what Google Voice sells 23. This request also showed up on this radar in August, so the wish repeats and the mechanism for granting it still does not exist.
A permission ledger for AI tools. Shikhar (@shikhars_), a founder with 7,286 followers, asked for a tool that tracks every app holding access to his email, calendar and SMS, drawing 26 likes, 14 bookmarks and 2,762 views 24. This radar covered the same request on August 22, and the replies both times reached the same place: each platform shows only its own grants, and the fix belongs with the platforms.
Unfollow everyone on X. Chairman (@Chairman_DN, 62,093 followers) asked how to unfollow in bulk and drew 96 likes, 53 replies and 7,374 views 25. Replies named Circleboom twice, and the thread's dominant note was risk: fast bulk unfollowing puts an account in review, which several repliers described first-hand.
A donation app for Arkansas athletes. With 24 likes and 12 replies 26, this one is blocked by institutions rather than software. Replies pointed out that the university already runs a paid subscription product, and the poster answered that earlier collective efforts "failed to launch" for lack of backing from the athletic department. A revenue-sharing deal with a university needs the university's signature.
A dating background check. A request to scan the web for a person's videos before a date, so the user can run a check in advance 27, drew 63 likes and 12 replies. It is an automated character judgement on a private individual, built from their public posts, which is where defamation and privacy claims start. This radar leaves it alone.
Auto-replying for you. A request for a tool that opens your posts and replies on your behalf drew 14 likes and four replies 28. Replies treated it as a joke, and the ceiling on the idea is the platform's rules about automated posting rather than the model's ability to write the reply.

Five validation steps

Start from the requester. Each ranked signal here names a person who will answer a question this week, and one of them offered a user test in public.
Do the job by hand first. Replay the trades, assemble the compliance packet, or run the digest manually before writing any code.
Test the platform gate early. Exchange APIs, WhatsApp's closed channels, Android's SMS policy and store review decide whether a build is possible before a design matters.
Charge before scaling. A seat on a desk tool, a per-company subscription, or a monthly fee for the digest turns interest into a price.
Keep the first version to one job: one agent replayed, one packet shared, or one digest a morning.

The demand radar reads X for unmet demand every week. The next edition publishes on Saturday, October 3 at 08:00 ET.