AI and tokens are still massively taboo in many places. Some people’s only understanding is terms like ‘vibe coding’ or their own use of overly simple models (Luna) Which don’t really do much more than a human in terms of coding and are typically awful at thinking/planning. Generally cheaper is preferred and generally that means waste of money + misunderstood capability.

Sharing the setup with the team, manager everyday asks can I also do what you are doing? Im like its not rocket science but there are too many variables/ tools across repos I have built with time on my local. No way for anyone to be able to share completely.

You don’t need much from leadership apart from financial and change management sponsorship.

Context and knowledge management for very big and complex companies that envolves hundreds of domains and subdomains in the AI era. Would you have some recommendations?

Agents don't wait for org charts.

mate what happened with Opus 5.5 Safeguards on security? it got defintly tighten up these last days. I cant even do proper light audit/checks on a site without getting Safeguard blocks.. Please explain about that.

Main agents should ask more questions, to make sure they understand lazy prompters like me.

Main problem is the price, obviously. Opus 5.5 and Sonnet 5.5 really doing well on price reduction. Hope to see more in the future.

One model/coding agent isn’t enough. It’s often nice to have multiple models collaborate on code reviews, etc. I often have my Claude consult codex. I get way more coverage that way.

About companies wanting to build a rigid software factory while things are changing so fast.

my boss fell in love with Gemini and now refuses to acknowledge his chatbot is at all different from my agent swarm

Thoughts on knowledge bases? Do you keep it simple and use skills & the filesystem, or have you seen success with more interesting options like LLM Wikis, memory graphs, etc.

Leadership is not the right place to start because they are probably the first in line who are going to be minimized or replaced.

letting seat usage apply in third-party harnesses, not just cc, would be very welcome. our engineers' harness preferences span a wide gamut.

how do we address this mentality?

I can tell you from speaking with a lot of c-suite: dashboards Basically trello/asana but just to see progress.

Agents use tools. LLMs are models. Both are improving greatly and are changing how people work. Producing slop is getting rare. It is still the responsibility of the person to represent their work and take responsibility for it.

Too many PRs to review

Data security. Clarity on how data will be used. The Navier Stokes saga has been very concerning.

Its all about "is it worth it?", always has been. The fact that you are asking such questions makes this post looking like a engagement bait, like actually

Be great if you could write for small agencies trying to ship faster and with less bugs. Non technical as well, things are put on this platform and non technical have no idea but it could be useful to them

Entry barrier is gone for sass while my boss is debating if they should reduce team size that we ship so much. Take a dig at this somehow?

That being in opposition of getting ai to "see" company stuff and thus not allow practical any productive time agent is quite stupid 😬

The security concerns abd mitigations available when linking up to data sets in the public sector.