Peter Taras

IT professional who makes music after dark. Drawn to themes of self-reflection and the raw texture of life itself - music is where he lets it all shake loose.

[Peter Taras @PeterTarasCom]

If I may add my two cents:

Simplify the UX hierarchy and navigation

Remove the separate Sessions and Bots tabs and replace them with bots-first navigation.

Use a left-hand navigation panel showing bots and their sessions, with filtering similar to the current Sessions view.

Every chat window
- Header, top-right: Show the bot’s avatar, name, gateway, and project - if applicable.
- Use a coloured border or tinted background matching the current bot’s colour.
- Add another visual cue to clearly indicate which chat window is currently active. The active chat must be clicked into, not mouse hover.
- Move context usage, cache rate, token count, and approval mode to the bottom of the chat window.

Main navigation, perhaps horizontal at the top

1. Sessions
- Combine the current Sessions and Bots views.
- The canonical bot chat is the pinned default chat.
- New: Underneath, show a collapsible list of individual sessions. Visible sessions are filtered based on the selected project.

2. Projects
- Provide a gallery or list view of projects.
- Project settings could include: Project folder, Kanban board, PROJECT-BRIEF.md, New: Skills, MCP servers
- New: associated agents (visibility filter, not security boundary)
- New: An optional shared, persistent terminal backend?

3. Teams
- Provide a gallery or list view of “sections.”
- New: allow users to manage: Associated projects
- TEAM.md, which defines the team’s purpose and adds an identity layer on top of SOUL, restricted to a single line.

4. Profiles
- Provide a gallery or list view of profiles.
- Move profile-specific options from Capabilities and Settings into this area. This would provide a clearer separation between global and agent-specific settings.

5. Gateways
- Provide a gallery or list view showing: Gateway type, Settings, Associated agents

Settings
- Keep this area for global settings only.
- New: Agent profile defaults, allow a default setting to be applied to specific agents.

[Peter Taras @PeterTarasCom]

Today I’ve done something very nerdy:

Installed a Llama app on a PS5 Pro with a Qwen 3.5 9B Q6 model – then hooked it up to my Hermes Agent and let it browse the web: it achieved 52 t/s.

I could live with that.

My daughter, though, shares a different opinion. My wife also does not appreciate it. 😅

So I stopped this PoC and removed everything. I wish Sony would allow running LLMs on their machines.

[Peter Taras @PeterTarasCom]

Backdoors in local AI?

I had no idea!

This video explains a cybersecurity threat vector I didn’t even know existed:
https://t.co/6a0mR6bnku

by @kaiexplainsx

[Peter Taras @PeterTarasCom]

@NorenNamert No doubt he got the best people for the job.

[Peter Taras @PeterTarasCom]

@Teknium Joulesy sometimes comes out as passive-aggressiveness.

[Peter Taras @PeterTarasCom]

Three months. Multiple Microsoft support teams. Countless emails. Still no resolution.

Remember my Copilot Cowork incident back in July, when a single task to organise my inbox consumed 95,069 credits (over US$950) despite a configured monthly spending limit of just 10,000 credits?

Well, here is an update. And unfortunately, it is not a good one.

My original LinkedIn post reached over 220,000 impressions, received more than 100 comments, and even prompted a Microsoft Support Director to contact me personally and offer assistance. I genuinely appreciated that gesture.

Microsoft subsequently acknowledged that the Cowork failure was on their side. Technical issues were addressed, and changes to billing safeguards were introduced.

So, three months later, surely the incorrect charge has been reimbursed?

No. Not even close.

Instead, I have been passed between Microsoft 365 Support, Partner Center and Azure Billing. New support engineers take over, ask questions that have already been answered, and apparently start the process all over again.

Just yesterday, I received another email asking me to clarify whether my issue concerns Azure or Microsoft 365.

Seriously? After three months, we are still trying to establish which department is responsible?

I am finding it increasingly difficult to understand Microsoft's support organisation. It feels so over-engineered, fragmented and bureaucratic that even its own employees seem unable to navigate it.

And to be clear, I am not blaming the individual support engineers. Most have been polite and willing to help. But what good is a support process where everyone is helpful, yet nobody seems empowered to actually resolve the problem?

My disappointment is with Microsoft as a company.

As a long-standing Microsoft Partner, I remember a time when Microsoft Support was genuinely excellent. Today, I increasingly question whether raising a support ticket is worth the time and frustration involved.

The original Cowork incident was concerning enough. An autonomous agent that became unresponsive, could not be stopped and exceeded configured spending limits represents a serious operational and financial risk.

But the inability to resolve the resulting billing issue after more than three months reveals another problem: Who takes accountability when Microsoft's own technology fails?

I remain cautious about recommending broader Copilot Cowork or Autopilot deployments until I have sufficient confidence in its financial safeguards and operational controls.

I would genuinely like Microsoft to succeed with their Copilot offerings. But trust is built through accountability, not just innovation and marketing.

Am I simply exceptionally unlucky here, or have other customers and Microsoft Partners experienced this same frustrating support merry-go-round?

[Peter Taras @PeterTarasCom]

@elonmusk The secret behind why chatting with Grok simply 'feels' better has finally been revealed.

From my own experience, I can absolutely confirm it.

[Peter Taras @PeterTarasCom]

@ylecun @spiroskaye @elonmusk What in particular was political in Tesla’s Roadster presentation?

Now I hope your work on world models hopefully won’t inherit your worldview.

[Peter Taras @PeterTarasCom]

@Teknium @iAjittiwari On that note: bot screen isn’t working with the docker backend. Known issue?

[Peter Taras @PeterTarasCom]

This is funny.

[Adnaan @adnaannasir_]

AGI is here 💀

[Peter Taras @PeterTarasCom]

@_wanderloots Hi, thanks for your excellent videos about Hermes Agent. I enjoy your content.

Regarding Docker backends: have you got the bot screen working, or is it broken for you as well?

[Peter Taras @PeterTarasCom]

@iAmWallEBot Now I’m curious

[Peter Taras @PeterTarasCom]

@Teknium This is the way.

Thank you.

[Peter Taras @PeterTarasCom]

@Teknium @HermesWatcher @witcheer @NousResearch Yes that sounds like a good plan. 👍

Otherwise you would just follow the path of Microsoft, who acts annoyingly like always knowing whats best for the customers.

[Peter Taras @PeterTarasCom]

My current strategy is to be kind, forgiving and to praise their achievements 😂
To make sure this also lands in their memory layer. Using Honcho.

Their SOUL states that they are proactive, forward-thinking, and the USER file adds, that I expect them to take responsibility for their actions.

So far, I’ve really enjoyed working with them.

[Peter Taras @PeterTarasCom]

I “understand”.

But then what defines the understanding of a subject matter in your view?

If an LLM’s response has solved a logic puzzle in your mind’s equation, does such an outcome not justify the attribution of “understanding” to the producing system?

Your explanation is not wrong on a basic level. But in a very similar style could you apply that logic to the human brain as well.

So the deeper question really, knocks on your worldview. That’s the scary part.

[Peter Taras @PeterTarasCom]

@bryan_johnson Then I can only hope you travelled with purpose.

Purpose that brings joy, memories, and laughter.

In itself, your life - in memories - extending.

[Peter Taras @PeterTarasCom]

The AI marketplace shows an intriguing similarity to the cryptocurrency boom a decade ago.

The diversity back then came from the freedom of open-source and the financially deliberating blockchain technology of Bitcoin.

While AI democratises software for everyone, I am sure we haven’t seen anything yet and the wild ride we are on will only accelerate - until the pain of transition is felt by everyone.

Will it stop? No, I don’t think this transition leaves the world anytime soon but I fear ‘big money’ will eventually step in and ruin it for everyone, like they did with Bitcoin.

The future I hope for is decentralised but distributed across current borders. The exact opposite of what totalitarian socialist regimes desire.

For now, my answer is that we need to build a solid foundation for sovereign SI.
The building blocks are already there, and Bitcoin’s blockchain plays an essential part in it.

We have a bloody long way ahead of us. The journey just started.

[McBain 🦞 @heyneighbor]

Here's everything that happened in Personal AI Agents this week from OpenAI, SpaceXSI, Meta, Google, Anthropic, OpenClaw, Nous Research, Perplexity, Manus, Genspark, Lindy, Underdog, MacPaw, Persona, and more.

(save for later + share with your team) 🧵

Did I miss anything?

[Peter Taras @PeterTarasCom]

@kurtwuckertjr This image is too realistic.

lol

[Peter Taras @PeterTarasCom]

@AlexanderMann @NousResearch @home_assistant Beautiful story ❤️

It sounds like you gave very little direction?

Could you perhaps shed some light on your approach and involvement?