Brian Cline

@bcherny Whew, thought nextDialog was replaced by claude code desktop... nextDialog gently nudges when a session needs attention and couples the terminal(s) with the project.

We've been working on this for a while. Can't wait to hear what you think

@bcherny Aw man, guess all those tmux projects just got displaced as well as my zen for Claude code

Anyone else having issues with Claude Code today? It's like pulling teeth to get it to actually do the work. Feels sluggish today. Using: Claude Code v2.1.101

Been running Claude Code's No-Flicker Mode all day. Really like the experience - no jumping - copy/paste seems to work better too! NextDialog can now tune your Claude Code sessions and has the toggle for no flicker. nextdialog.io thanks @bcherny !

@janardhanp22 Yes, I'm working on a "hover" session-idea - this would allow you to see the last 3 lines

Just shipped "Catch Me Up" for Claude Code. When you're juggling multiple sessions, it's easy to lose context. Now you can pull down a timeline on any session and instantly see every file edit, bash command, and tool call - organized by turn. Scroll all the way back to the beginning.

It uses Claude Code's hook system. A local HTTP server per session captures events (file edits, bash runs, tool calls) in real time as they fire. No polling, no log scraping. Each session gets its own event ledger so there's zero bleed between projects. The timeline is a pull-down overlay on the session card. I use this daily to run about 5 active Claude Code sessions. And it's not just the sessions you're tracking. Each one usually has a companion terminal running tests or tailing logs. That's 10+ terminals before you know it.

@bwcline the session context problem is real - once you have more than 2-3 active projects the bleed becomes a real tax. how does it pull the session data - does it hook into claude codes internal logs or does it need its own tracking layer?

.@rgb_prithvi's new post on harness design for long-running agents is worth a close read. The finding that criteria in the generator's prompt improved outputs before the evaluator loop even kicked in really stood out. I've been exploring a similar angle building a Claude Code session manager: heavy investment in anti-pattern libraries that make the generator's first pass much stronger. Curious how far pre-loop criteria tuning can go on its own before the external evaluator becomes necessary. 70/30? 50/50? anthropic.com/engineering/ha…

I run 5-10 AI coding agents in parallel. The bottleneck isn't the tools — it's my brain trying to track all of them. BCG found productivity drops after 3 concurrent agents. Boris Cherny says the job is now "context switching." Nobody's designing for that.

@bcherny I saw you run 10-15 Claude Code sessions in parallel. How do you deal with the cognitive overload of context-switching and unblocking that many sessions? Any patterns that keep it sustainable?

Everyone's asking if AI will take your job. Almost nobody's asking what you could do with it that you couldn't do before. Klarna replaced 700 agents with a chatbot, celebrated the savings, then rehired humans within a year. That's not an AI failure — it's a design failure. We keep optimizing for the spreadsheet instead of the person. I wrote about why the better question changes everything. @brianwcline_32237/what-if-we-have-it-backwards-3ceb1016ffeb" target="_blank" rel="nofollow noopener">medium.com/@brianwcline_3…

Enterprise teams can't even validate AI for sales forecasting — they don't understand the model well enough to know when it breaks. Now we're deploying non-deterministic systems in military contexts and telling the only people who understand the failure modes to step aside. That's the actual story.

That's the regulatory gap, yes. But there's a deeper technical one: these are non-deterministic systems. The model makers are still finding edge cases in controlled settings. Nobody outside Anthropic or OpenAI can fully evaluate how these models behave at the margins. We didn't just lose a company willing to say no — we lost the only people in the room who understood where it breaks.

Anthropic just walked away from a $200M Pentagon contract rather than remove safety checks from Claude. Hours later, OpenAI signed their own deal. The question nobody's asking: why were the safety checks negotiable in the first place?

<strong>Oracle Crushes Q2: 5 Reasons Behind 43% Cloud Surge</strong> accelerationeconomy.com/cloud-wars/ora…

LIVE on #Periscope: Behind the scenes at the Salesforce Lightning Launch party! pscp.tv/w/aK0CDDgxNTkw…

@Benioff As one of your largest Indiana based customers, Angie's List supports you. Thank you for taking an interest.