Cipher Projects

AI agents & cloud engineering for Singapore & Australia

[Cipher Projects @cipherprojects]

Instead of worrying about the dev jobs going away, look at the role that's growing.
#forwarddeployedengineer #aiagents #aiimplementation

[Cipher Projects @cipherprojects]

Saturday 31st Oct // Ho Chi Minh City 🥥

Workshop focussed on adapting the strategies and frameworks used by enterprise FDE teams and applying them to helping small to medium business deploy AI.

10-12am @ HOFF Specialty, Vo Thi Sau

[Cipher Projects @cipherprojects]

This weekend we supported the @spacexai Grok Bot Meetup.

Every person had a different background but they all want to learn to automate their life and business tasks with AI.

Amazing to meet everyone and thanks to @frontierclubx and our Founder @techopsasia for connecting.

[Elon Musk @elonmusk]

Starlink

[Keith Vaughan @techopsasia]

One control panel for a human life. Interesting feature.

Cursor made this for me already but I have expanded the model beyond coding.

All the great designers are converging on a few things with both Cursor and Grok Bot and most people who use Claude Code CLI building hand-rolled solutions.

Cursor today released "Projects" where you talk to one long running thread that handles all sub agents and task rather than flicking between heaps of chat sessions. Grok Bot already made each agent have a long running persona and this is a PM persona on top of Cursor.

Interesting the most because I built for myself already on top of Cursor CLI agents.

It's the product manager model, not the multi-CLI panel with many persistent TMUX sessions or disjointed chat sessions.

When I do coding task I talk to my project manager who owns my business knowledge. It assigns the tasks to Cursor CLI subagents which run on incus containers (written in go). It also works on my phone via voice and is secured with MFA and some other secret tricks.

But mine has one difference.

So I only ever talk to Phuong my product manager. She handles the tasks, and she monitors my whole life.

I manage my fasting, meditation, martial arts sessions and weight lifting per week. Macronutrient profiles and calories of the food I eat each day all tracked over the month.

One control panel for a human life built by Cursor via SSH on my VPS on top of hermes now edited by the hermes itself as I ask for a new panel.

I also believe this will be the norm, most people will soon just talk and have their own personal dashboard appear built around 20 different API's, some of those paid for by their agent's monthly wallet balance.

Since the AI boom most of us have been working 80 hour weeks, it's time to at least reclaim our walking time and talk to our agent on the go with our voice.

--

Views are my own not of @cipherprojects

[Cipher Projects @cipherprojects]

Saturday was Frontier Club AI Meetup #3.

Sarthak Langde walked through using local models to lower LLM costs and there was a lot of good energy.

Cipher Projects was glad to sponsor again with Yobeve and Your Event Kit.

@sarthaklangde

[Cipher Projects @cipherprojects]

Is AI Killing Indie Hackers? Levelsio vs Greg Isenberg

Pieter Levels sees indie revenue falling. Greg Isenberg sees a record startup boom. Both are reading the same market. Here is how that works, and what to do about it.

https://t.co/WcGVY6NDqj

[Cipher Projects @cipherprojects]

@sofrain9 It depends on the project. We are not tied to any framework or cloud.

[Cipher Projects @cipherprojects]

We come from platform and cloud infrastructure, so the agents and software we build are production-grade, secure, and compliant by default. Not bolted on after the fact.

[Cipher Projects @cipherprojects]

Head over to https://t.co/UW6QJExf5P

[Cipher Projects @cipherprojects]

AI agents can write documents.

They should be able to send and manage signatures too. And that's why we built SumoSign.

[Cipher Projects @cipherprojects]

https://t.co/nxMVRxjAcJ

[Cipher Projects @cipherprojects]

https://t.co/hYgpdY2VY5

[Cipher Projects @cipherprojects]

https://t.co/Jg8pbeQ9JB

[Keith Vaughan @techopsasia]

Claude Fable 5 doesn't stop at the first plausible answer.

It keeps testing until there's real proof the task is done.
We can't recreate that in GLM 5.2.

But we wrote a free system prompt that pushes it toward the same loop: plan once, then act, verify, retry.

Try it: