Replit AI — Module 22.6

How This Lesson Fits the Module & Volume

Bolt.new and Lovable generate apps in specialized builders. Replit AI sits in between those builders and desktop IDEs: a cloud development workspace (multiplayer editor, run, DB, hosting) plus AI—historically Ghostwriter, now headlined by Replit Agent / Assistant. It closes Module 22.6 because it already deploys (Vol. 18 gravity) and hands off naturally to Module 22.7 automation: once an app is up, n8n / Make / Zapier wire the rest of the business.

Vol. 21 coding assistants + Vol. 15 agent loops + Vol. 18 deployment all show up in one vendor. Education and collab are first-class—unlike Cursor’s solo desktop default.

Learning Objectives

By the end of this lesson, students should be able to:

Describe Replit as cloud IDE + run + host + AI Agent/Assistant, not merely Tab.
Contrast Agent (agentic build/deploy) vs in-editor autocomplete/assistant.
State privacy/residency implications of a cloud workspace vs local Cursor.
Compare Replit to Bolt, Lovable, Copilot, and Claude Code qualitatively.
Use Replit for education/collab/spikes; graduate serious systems to git + your Vol. 18 pipeline when needed.
Bridge 22.6 → 22.7: apps still need automation graphs (n8n/Make/Zapier).

Definition

Replit AI refers to Replit’s model-backed features inside Replit’s cloud IDE: an Agent that can scaffold, modify, run, and help deploy a Repl, plus chat/assistant and (historically) Ghostwriter-style completion. The platform itself is a hosted workspace: files, language runtimes, secrets manager, collaboration, and one-click hosting. Treat Agent as Vol. 15 HITL (review before public deploy). Product names and credit meters evolve—confirm the current Replit AI docs and plan sheet.

Cloud IDE: Autocomplete, Agent, and Deploy

Surface	Pattern	HITL bar
Completion / Assistant	Autocomplete + chat in the cloud editor	Accept snippet / review edit
Replit Agent	Agentic multi-file + run loop	Review plan, secrets, and deploy checkbox
Hosting / Autoscale-class deploy	Vol. 18 “platform as runtime”	Env secrets, domains, access control
Multiplayer	Human–human collab with AI in the room	Who may click Deploy?

Replit vs Bolt

Realer runtimes (many languages)
Not limited to WebContainers
Heavier cloud workspace model

Replit vs Lovable

Less “pretty CRUD opinion” by default
More general-purpose student/dev VM
Deploy is first-class on Replit

Replit vs Cursor

No local offline-first default
Collab + classroom superpower
Agent can reach hosting—higher blast radius

Privacy and Repo Context

The “repo” lives on Replit’s cloud unless you mirror to GitHub. Indexing and Agent reads happen where the workspace is. That is convenient for schools and a procurement issue for enterprises (residency, SSO, who can fork a Repl). Use Replit Secrets—never hardcode keys in Agent prompts. Public Repls are public. Vol. 20 privacy: student PII in classroom apps needs the same dummy-data rule as Lovable.

Catalog Snapshot (Qualitative)

Dimension	Replit AI	Cursor / Copilot	Bolt / Lovable
Pricing posture	Core/Teams-class subscription plus Agent/compute credits—live sheet	IDE seats + premium/agent usage	Builder credits + seats
Strengths	Cloud IDE + run + host + collab + Agent in one; education; many languages	Local/enterprise IDE depth; GitHub/JetBrains gravity	Fastest pretty web MVPs / WebContainer spikes
Weaknesses	Cloud gravity & residency; Agent deploy blast radius; not a full replacement for mature CI/CD estates	Weaker zero-install classroom story	Narrower runtime / stack opinions
Agentic vs autocomplete	Both (Agent headline + editor assistant)	Both (Tab heritage or Agent)	Mostly agent-first scaffold
Privacy / repo	Cloud workspace + Secrets + org/edu controls	Local disk + vendor prompt path	Prompt + generated project in vendor cloud
Vol. 18 deploy	In-platform hosting or export to your Docker/K8s	Your CI	Export then CI

Agent Deploy Gate Sketch

# .replit / Replit Secrets — conceptual (UI names evolve)
# Secrets: DATABASE_URL, SESSION_SECRET  (never in chat logs)
# run = "python -m pytest -q && python main.py"

AGENT_POLICY = {
    "may_edit_files": True,
    "may_run_tests": True,
    "may_click_deploy": False,   # human only (Vol. 15 HITL + Vol. 18)
    "public_repl_ok": False,     # classroom dummy apps only if public
    "pii_allowed": False,
    "forbid_exploit_tasks": True,
}

def agent_turn(action: str, human_deploy: bool) -> str:
    if action == "deploy" and not human_deploy:
        return "deny: human must deploy"
    if action == "write_secret_to_source":
        return "deny: use Replit Secrets"
    return "allow"

# After Module 22.6: wire the running app to n8n/Make/Zapier (22.7)
# e.g. webhook on new signup → automation graph, not more Agent autonomy.

When to Pick Replit AI

Pick Replit when

Education, workshops, multiplayer pairing.
You want run + host without local Docker (yet).
Polyglot spikes beyond WebContainers.
Agent-to-preview is the teaching loop.

Pick something else when

Air-gap / strict residency → local Cursor + self-host CI.
GitHub Enterprise + JetBrains → Copilot.
Pretty CRUD for PMs → Lovable.
In-browser JS only → Bolt.

Related Lectures

Lecture	Role
Cursor / Copilot	Desktop/plugin daily drivers vs cloud IDE
Coding assistants	HITL metrics still apply
Agent loop / HITL	Agent deploy gates
Deployment	Replit host vs your K8s
n8n (22.7)	Next module: automation after the app exists

Common Misconception

“Replit Agent should auto-deploy every successful test.” Deploy is a Vol. 18 + HITL act. Second: public Repls are a privacy feature. Third: Replit replaces GitHub Enterprise for every regulated shop. Fourth: Ghostwriter-era autocomplete is the whole product. Fifth: inventing credit prices. Sixth: Module 22.6 ending means automation is unnecessary—22.7 starts next.

Knowledge Check

Short Answer: What three platform jobs does Replit combine with AI? Answer: Cloud IDE / run, collaboration, and hosting (plus Agent/Assistant).
True/False: Replit Agent and editor autocomplete are the same HITL bar. Answer: False—Agent can reach run/deploy; completion is snippet-level.
Multiple Choice: A privacy implication of Replit vs Cursor is: (a) workspace lives in the vendor cloud by default, (b) no models are used, (c) air-gap is automatic. Answer: (a).
Short Answer: Where should API keys live on Replit? Answer: Replit Secrets (not source or Agent chat).
True/False: Replit is usually stronger than Bolt for polyglot / non-WebContainer runtimes. Answer: True.
Multiple Choice: Pricing posture is: (a) workspace plan + Agent/compute usage, (b) a fake $7 figure, (c) only Zapier tasks. Answer: (a).
Short Answer: Why does 22.6 end on Replit before n8n? Answer: Replit already deploys apps; 22.7 automation wires those apps into business graphs.
True/False: Classroom public Repls may contain real student PII. Answer: False—dummy data only.
Multiple Choice: Who should click Deploy in the sketch? (a) human, (b) Agent always, (c) a free internet stranger. Answer: (a).
Short Answer: Next module lecture? Answer: n8n (Module 22.7).

Key Takeaways

Replit AI = cloud workspace + Assistant/Agent + hosting—education and collab superpower.
Separate autocomplete from Agent; gate Deploy with HITL.
Cloud residency and public Repls are privacy decisions, not footnotes.
Use Replit for spikes/classes; graduate regulated systems to your Vol. 18 estate when required.
Next module: n8n — automation platforms.

Trainer’s Guide

Capstone 22.6: Teams build a dummy webhook receiver on Replit (or export). Agent may write code; only humans deploy. Then preview 22.7: that webhook becomes an n8n trigger.

Exit ticket: One sentence: when Replit vs Cursor vs Copilot vs Lovable. No dollar figures.

Recap: Replit closes AI IDEs by adding cloud run, collab, and deploy to the agent/autocomplete story. Continue to Module 22.7 n8n.