Git worktrees makes this possible. I often have 4+ Claude Code’s going usually working on different aspects of the codebase (testing, documentation, research, feature work) then review the PR’a as their posted
2026-03-24 17:27:29
85
Lord Kaizer 🍀 :
Most people are just larping or just don't care 😅😅 one AI agent is way too fast to keep up with and review, 7 in parallel is pretty much a joke
2026-03-24 17:18:11
195
LB :
How did dev teams work in parallel?
2026-03-25 02:48:20
14
aromapex :
god, I am so glad to see someone else say this. I haven't for the life of me been able to figure out how the hell people have multiple agents doing shit simultaneously. like how are the dependencies between components handled when so many changes are occurring at once. and how are the changes even audited... like there do u ensure all that interlinked code is actually safe/scalable/well-thought. the only way I can see this is if the feature each agent is working on is extremely isolated. but those are so rare... as u say, I'd also love to see an actual example haha
2026-03-24 17:19:22
73
rale :
Max is like 4. 1 for the current ticket, 1 for side project to help team, 2 for debugging customer issues. But all of it is completely independent
2026-03-24 18:53:10
15
JaZahn :
Well architected systems with atomic functionality?
2026-03-26 21:07:48
5
WafiToffy :
They’re writing bad code / code that has no consequences
2026-03-24 19:04:43
22
Hritik Mehta :
6-7 parallel features is wild ngl, how modular even is your codebase and how fast do you context switch to be able to manage all that???? At max I’m communicating with 3 agents at a time where 2 are planning a feature by identifying gaps in plans back and forth while maybe another inferior agent is making some quick UI changes while the bigger agents are running investigations
2026-03-24 17:25:56
6
random_inferences :
git worktrees + get shit done plugin or super powers plugin. gsd and superpowers require more setup but then you let it off and run for awhile. context switch your brain is the hardest part since the worktrees handle code isolation
2026-03-25 15:59:06
5
Evan Braun :
I partially agree. What I’m doing is using spec driven development to create a spec and then I ask the main driver agent to understand the spec/codebase and then split the work into tasks that can be developed in parallel based on X agents that I’m spinning up.
2026-03-24 17:26:12
13
Will :
Because they aren’t. 99% of all the AI stuff you read on X is BS and isn’t being done by real developers.
2026-04-14 16:09:31
1
user75478864432190 :
who said anything about good performant and a good UX lol
2026-03-24 18:33:42
8
War Badger :
Tbh it depends on context. Is the thing you’re building big and understood? Spend a day building a project context to be split into pieces. “Port this C project into Rust”.
2026-03-26 18:21:30
0
Amber Bennoui :
Same here - how do you review 6 massive changes at once? Are you testing? Are you making sure that the agent is following your requirements? It can barely handle one task I give it and I’m a PM with insanely specific requirements lol
2026-03-28 01:51:36
1
Jonathan banks :
I’m not building 6 different features, I’m building 6 different companies, that’s why I’m broke and none of them make money
2026-04-15 13:49:10
1
Kyngston :
you spec’d the features before starting implementation
2026-03-26 06:58:37
1
user3817829366041 :
People trying to posture as AI early adopters. 🤦🏻♂️
2026-03-26 06:32:30
4
VP350 :
Large rewrites or making the code more modular usually can use up quite a few agents when using superpowers. Since all new code is modular the agents can write the code in parallel and save time. Also have a couple agents on writing unit tests and reviewing the code
2026-04-19 18:27:58
0
Kesh :
Highly unreliable doing this with Claude code. Codex can one shot features end to end so if you spend a ton of time on the plan, you can run them all on worktrees.
2026-03-27 12:10:35
0
Lucid Reverie :
Hi. Im doing this..I run 16 windows on 2 pcs... building kernel up os and a 64 layer ai..
2026-03-26 07:36:39
0
Egyptian Chef :
I just gave a talk about this at SXSW. Claude is ingesting feature requests from notion and working through them. Multiple agents with specialized skills working through merge conflicts. Human in the loop to approve the prod pr merge
2026-03-25 03:23:16
2
CaddieAI :
Yep….
2026-03-24 19:15:18
2
giugfsgu :
Integration (playwright) and unit tests are the only instances I found where that kind of parallelism is not a hindrance. Also UX assessment/critique with playwright mcp can be scaled horizontally.
2026-03-25 07:44:05
0
tg :
I can’t keep more than 2 sessions active without loosing mental context. If the orchestrator decides to run parallel scoped agents that’s great, but still one feature per project at a time.
2026-03-24 19:55:55
2
lcmilstein :
Don’t you have teammates? How DOES your manager do it?
2026-03-24 23:03:27
1
To see more videos from user @evisdrenova, please go to the Tikwm
homepage.