x-octo home Business judgment on AI products
中文

VOL.2026.10.18 Today's call 3 min read

Agents are leaving the chat box for specific jobs: ad ops, portfolio construction, demo recording and scientific modeling are being broken into executable actions.

Sunday, October 18, 2026

—
Ad monitoring and launches are being split into deterministic actions an agent can call, with adsctl entering daily operations from the command line.
—
Data-heavy back-office finance work such as portfolio construction is starting to be taken over by agents; Rowspace claims days shrink to minutes, but humans still confirm.
—
Banks collectively took defensive measures against consumer-facing AI agents, adding compliance and integration cost to autonomous agents in regulated settings.
—
AI comic dramas entered super-long serialization, with 41 series IPs and top-season revenue around RMB 1–2 million, shifting competition to IP and throughput.
01

Today's positive direction

4 picks

The clearest signal today is that agents are leaving the general chat box and entering specific jobs with defined input material and defined deliverables. Ad monitoring and launches, portfolio construction, web demo recording and biology model building are all being broken into actions an agent can call. This is not another batch of ad dashboards or analytics panels; it is pulling the repetitive steps out of old workflows and handing them to an agent.

01

adsctl

For ad operations staff managing Meta ad accounts: account monitoring, insight reports and campaign launches are handed to a command line or AI agent, which runs deterministic monitoring, generates reports and launches rollback-capable campaigns, so users get reports and live campaigns. The notable choice is the path of deterministic actions plus rollback rather than rebuilding an ad interface; how the risk boundary around launches is drawn will decide whether operations teams truly trust it.

02

Rowspace

For investment analysts: during portfolio reviews or new allocations they previously had to manually gather holdings, risk and market data and spend days assembling a proposal; Rowspace's agents are said to take these materials and generate portfolio construction results, compressing the cycle to minutes, with final proposals still confirmed by investment staff. It targets data-heavy, repetitive back-office finance work rather than a Q&A assistant; accuracy and the depth of human confirmation still need verification.

03

cursorcam

For developers who need to demo a web app to clients or teammates: an AI agent clicks through the page and returns a playable MP4 demo that zooms in on each click; it also runs standalone as a command line tool. It replaces manual screen recording and editing with an agent that executes and delivers the finished video, and the deliverable being the video itself makes the payment trigger relatively direct.

04

Anima

For biology researchers or teachers: describe a model in plain English and AI generates an interactive biology model for explanation or exploration, so users get an operable model rather than a static diagram. It turns diagram-based explanation into operable simulation, angled toward teaching and science communication; input scope and accuracy still need verification.

02

Market context

The banking shift deserves to come first: in October 2026, Citi, Bank of America and JPMorgan collectively took defensive measures against consumer-facing AI agents. This shows that in regulated finance, autonomous agent access is constrained by institutional risk controls, and deployment carries extra compliance and integration cost, shifting competition toward trust and auditability. This contrasts with products like Rowspace that serve internal investment workflows: internal back-office processes and consumer-facing agents face different risk thresholds.

On content, AI comic dramas entered super-long serialization, with series IPs rising to 41 and top-season revenue around RMB 1–2 million. Scalable serialized production and monetization have taken shape; as production barriers fall, competition shifts to IP and sustained throughput.

On tooling, Anthropic's internal report says agent working hours reached six times human hours; this is a single company's internal statistic and only a reference for usage intensity, not evidence of overall substitution. Apple plans to bring its first AI features to the China market in December 2026, meaning a distribution entry with a large installed base may gain system-level capabilities, which is both pressure and a channel for the application layer.

03

To verify

AIsa has only the fact of over $9 million in funding, with no visible step of old work being replaced; Craft Agents offers only a tagline; CrewTower Mod for Claude Code is a single sentence and an interface plugin tied to one model, with limited moat. The Benefit Street Partners case shows a general coding agent borrowed by a non-engineering team to compare credit agreement clauses and valuation data, useful as evidence that agents are entering structured-text work, but it does not state time saved.