What did @samuel_wong_ post on X, week of Sep 15 – Sep 21, 2026?

The week was dominated by enterprise AI architecture and governance — @samuel_wong_ kept returning to the gap between buying AI tools and actually transforming the business around them, with his most-read post arguing that nearly half of agentic projects are heading for cancellation because organisations skip governance and jump straight to autonomy.

Weekly brief · week of Sep 15 – Sep 21, 2026

Get this account in TelegramFollow

AI Business Strategy & Architecture

~43%

Across six posts, the through-line was clear: organisations buy AI tools but refuse to change the business models those tools are meant to disrupt, and the governance layer that makes agents safe is the part everyone skips. [1]

BI migration and semantic foundations: When switching BI tools, migrate the logic, not the layouts — governed metric definitions become the context that makes AI analytics actually work, while 80% of dashboards were never more than one-off answers dressed up as permanent fixtures [2]. The hardest part of agentic analytics isn't the AI; it's getting everyone to agree on what "active user" means, because without that shared semantic foundation every AI tool just automates the confusion faster [3]. Rolling out AI internally should be treated like a product launch: you need internal marketing, training, and champions on every team [3].

Agent governance and infrastructure: Agent-native infrastructure is about how agents interact with software, why long-running tasks break, and what safer execution requires [4]. AI agent governance must happen at runtime, and governance changes fundamentally in the face of agentic speed [5]. His most-discussed post — 136 views — argued that enterprise AI agents deliver measurable ROI when you match the right maturity level to the right process and govern it properly, but most organisations skip governance and jump straight to autonomy, which is why nearly half of agentic projects are heading for cancellation [1].

Business model transformation vs tool-buying: "Titans don't fall from lack of technology, they fall from lack of imagination," and most companies are buying AI tools while refusing to transform the business model those tools are supposed to disrupt [6]. The question of where the graph should live — whether a serious AI system needs a graph database — has a different answer for every workload [7].

other

~21%

A photo post captioned "Late to the party" with no alt-text and no substantive claim [8]. A mentorship link framed with the thought that mentorship is about having someone watch you play and notice the thing you've been doing wrong for years — hard to ask about something you haven't noticed [9]. A hobby-AI article about what people make when nobody is counting the hours, followed by a self-deprecating quip about his 2,056 GitHub contributions [10].

Curated Industry Reads (~29%): Shared four external links with light framing — a geo experimentation primer for measuring marketing incrementality [11], LinkedIn's Engineering CTO on building AI-native products in 2026 [12], a piece on how two SpaceX/AI designers use Grok Bot to build personal sites and prototypes [13], and a Substack on where graph databases should live in AI systems [7]. The LinkedIn and Grok pieces skew toward practitioner insight; the geo-experimentation link is a methods reference for marketing measurement.

AI-Driven Work & Startup Dynamics (~7%): The low-hanging fruit in software is gone — what's left is the messy, undocumented work that can only be understood from inside the customer's walls, and that's why everyone's suddenly "forward deployed" [14]. The point connects to the broader week-long theme: the hard part of AI isn't the model, it's the organisational and semantic messiness that surfaces once you try to deploy.

1 replies · 136 views@samuel_wong_ pushed back on the enterprise-agent hype cycle, arguing that measurable ROI only shows up when you match maturity to process and govern properly — and that skipping governance for autonomy is why nearly half of agentic projects are heading for cancellation. [1]

Get this account in Telegram, free.Mirra posts every new brief to a private channel. Joining a channel that already exists costs nothing.Free to join · no X or Telegram credentialsFollow this account

How Mirra writes a brief

Three steps, the same way every time — so a brief can be checked rather than trusted.

1
Every public postThreads, replies and quotes the account published inside the cycle window. Retweets are read for context, never counted as authorship.
2
Grouped by topicMirra weighs how much of the window went where, then writes one summary per topic — the dominant topic gets subgroups.
3
Nothing unlinkedEvery claim and stated position points back to the post it came from, so you can check it in one click.

Questions

A summary of everything one X account posted inside a cycle — grouped by topic, with a citation on every claim back to the post it came from. Mirra writes it; the account does not.

Once per cycle. A weekly brief is published at the end of its seven-day window, and this page then shows the newest one.

Yes. Add any public X account and Mirra starts covering it — you get briefs in the dashboard, a private Telegram channel or by email, on the schedule you pick.

Mirra emblem

Follow the accounts that matter. Skip the feed.

The algorithm hides posts. You both lose. Mirra reads every public post, groups what was said, and delivers it on your schedule.

Start free trialBrowse public briefs

This is an automatically generated summary of @samuel_wong_'s public posts on X — the linked originals are the source of record. Mirra is not affiliated with @samuel_wong_.

Are you @samuel_wong_ and want this page removed? Email support@mirra.to.