AI is not your buddy, watch out.

[Non Sequitur]

It can, just download superwhisper or any of the other whisper variations.

Would you recommend a microphone too? I'm thinking about buying a condenser microphone hoping it will clean some of the dictation.

 

[cashcatz]

Would you recommend a microphone too? I'm thinking about buying a condenser microphone hoping it will clean some of the dictation.

I just use built in laptop mic, the AI dictation is good at figuring it out. But honestly I prefer typing over dictation (I'm a fast typer and I get tired from speaking). All I can suggest is try superwhisper - the main thing is to understand the keyboard shortcuts to activate it.

 

[cashcatz]

Are AI tools reliable enough to be used at in commercial settings?

Absolutely not, and the people on Facebook encouraging appraisers (who are generally not technical) to set up OpenClaw are borderline unethical.

 

[cashcatz]

redacted - link won't post.

 

[NP_MAI]

Are AI tools reliable enough to be used at in commercial settings? If so, should they be given “autonomy” to make decisions? These are the questions being raised after at least two internet outages at Amazon’s cloud division were allegedly caused by blundering AI agents, according to new reporting from the Financial Times.

It depends on what you are using them for. There are areas where AI can automate menial tasks across our profession and I think the residential folks who embrace it will see massive near term benefits, but I am worried about the long-term impact to residential folks.

Data privacy remains a significant concern and clawdbot should not be given free reign. This is flat out not safe to run unless you take significant precautions (dedicated hardware, network lockdown, sandboxing,etc.).

Claude cowork is great as is their excel add in and the simple web-based chat. It also has full integration with google drive, sheets, and docs, and you can build some really cool agents to run things within your business. I built an email agent over the weekend that integrates with google sheets and manages my team's job logs, reaches out to property contacts when a job comes in, sends other necessary emails and follow-ups and its all automated. Claude excels in reading leases and I am building a couple of agents right now to read leases, provide lease abstracts with page references, reading of CAM recs, and preparation of an importable rent roll.

This is only scratching the surface and things are changing on a daily basis.

 

[cashcatz]

I built an email agent over the weekend that integrates with google sheets and manages my team's job logs, reaches out to property contacts when a job comes in, sends other necessary emails and follow-ups and its all automated.

This is neat, can you share more about your setup?

 

[Vermonter]

This is neat, can you share more about your setup?

Interested to know what he's using for token efficiency on something this complex. Cowork can get expensive quick. I know there are paid applications out there, but that kinda defeats the purpose.

https://bolt.new/ looks interesting

There's also this https://github.com/jgravelle/jcodemunch-mcp

Currently using opencode cli with a mix of local and paid models, but it's not officially supported by Anthropic. I tend to use the paid models for complex queries and the local models for simple stuff. Qwen3.5 dropped and is impressive so far. 35b struggled but 27b works well on my hardware.

Tend to struggle settling on a workflow but opencode seems to have set me straight.

 

[NP_MAI]

This is neat, can you share more about your setup?

Testing locally this weekend before deploying to Google Cloud. But its Python-based and monitors a dedicated gmail account, extracts job data, writes it to a google sheets tracker and then generates a number of templated emails to: the property contact (extracted from LOE), appraiser (with clarification questions on scope), and offshore help (instructions on scope driven by appraiser responses). It's set up to follow-up automatically at varying intervals if responses aren't received and will escalate to me if responses aren't received after varying time periods.....hopefully this provides some idea of what it is. Right now, the setup dead ends in sending an emails to me that I will manually forward until testing is complete.

 

[NP_MAI]

Interested to know what he's using for token efficiency on something this complex. Cowork can get expensive quick. I know there are paid applications out there, but that kinda defeats the purpose.

https://bolt.new/ looks interesting

There's also this https://github.com/jgravelle/jcodemunch-mcp

Currently using opencode cli with a mix of local and paid models, but it's not officially supported by Anthropic. I tend to use the paid models for complex queries and the local models for simple stuff. Qwen3.5 dropped and is impressive so far. 35b struggled but 27b works well on my hardware.

Tend to struggle settling on a workflow but opencode seems to have set me straight.

This is not token heavy, its actually pretty light. The lease analysis that i'm working on will burn through credits....so it'll be interesting to see how it shakes out.

 

[Non Sequitur]

I know OpenClaw is all the rage right now, but IMO Claude Cowork is the way to go.