Replit founder Amjad Masad has moved away from conventional prompting toward short outcome-focused prompts that kick off orchestrated loops. These loops run parallel agents, apply a computer-use verifier for iterative feedback, and dispatch specialized agents to handle security checks, production fixes, and SEO tasks. The approach mirrors capabilities introduced in Replit Agent 4 earlier this year.
I’ve been doing “loops” for a while now. I don’t do much traditional prompting. Most of my prompts are barely a sentence expressing an outcome.
- my orchestrator prompts parallel agents - my computer use verifier gives it feedback - my security, production, and SEO agents generate prompts for fixes The industry is typically 3-6 months behind what we’re doing at Replit.
@amasad do you use loops?
Masad claims the workflow places Replit three to six months ahead of typical industry practice, though measurable speed or quality gains are not detailed in the post. A computer-use verifier supplies the key feedback loop that traditional prompts lack.
No specifics on the models powering the orchestrator or verifier appear in the available posts, leaving open how widely the same loop structure is used by Replit customers versus internal teams.
Positive users praise Replit's multi-agent loops and feedback loops as the future of agent orchestration, while negative users dismiss the approach as an old buzzword for automation.
@amasad Will you share your loop details? I’m putting together a library of loops
I’ve been doing “loops” for a while now. I don’t do much traditional prompting. Most of my prompts are barely a sentence expressing an outcome.
- my orchestrator prompts parallel agents - my computer use verifier gives it feedback - my security, production, and SEO agents generate prompts for fixes The industry is typically 3-6 months behind what we’re doing at Replit.
@amasad How loops is suddenly gaining traction ? Did I miss something
@amasad I talk about this all the time. Replit has been doing loops for a long time. When I teach loops, I specifically show how Replit does it. You guys are leading the way.
@amasad Do you have any human level review before the action or change is accepted?
@amasad what are you doing now and what you think about doing and trying next
@amasad "barely a sentence" works when the verifier is doing 80% of the steering
@amasad try using Lore to improve context across agents, keeps them grounded in YOUR truth instead of relitigating.
https://github.com/tcballard/requirements-as-code?v=3
@amasad Very true and was obvious if people paid attention to the strategy you all implemented with agent 4
Beautiful
@amasad Enable Feedback Widget to Replit agent was my intro to them; got me into loops a few months before it was mainstream.
This is exactly how we're trying to architect real-world B2B systems right now. While everyone is stuck building traditional chatbots, we have parallel agentic loops autonomously catching missed leads, parsing CRM data, and dispatching tradesmen in the background. The biggest unlock right now isn't just code generation—it's deploying these loops to solve physical $80k/yr business bottlenecks.
@amasad yea
@amasad Can you give a detailed example on the same lines for traditional prompting vs a loop
@amasad So the prompting didn't go away, it just got distributed across your verifier, orchestrator, and feedback loop? That's not fewer prompts, that's prompt tax with extra steps.
@amasad lol you are delusional if you think ur the only one at the frontier or beyond :)) - get ready to be distrupted big time , platforms like Replit and loveable will soon be eaten alive - the only way to survive is to beg the LLM providers to provide you tokens at discount .
@amasad
@amasad You’re not prompting the orchestrator?
@amasad Man I have not started using /loops I need to start learning how to use it
@amasad Invitation open to my upcoming podcast.
@amasad I've been watching and trying to learn what makes these kinds of loops work.I can't read the code but the output tells me it's doing something.
@amasad 3-6 months is a very brave claim