Gitpod's pitch was simple enough that it changed expectations permanently: open a URL, get a running dev environment with the repo cloned, dependencies installed, and services started. Once a team has that, going back to a README with fourteen setup steps feels like a regression.
Choosing a replacement is harder than it looks, because "cloud dev environment" covers at least three different products. Some are IDEs that happen to run remotely. Some are machine provisioners with an IDE bolted on. Some are programmable sandboxes with no IDE at all, which turns out to be what you want when the thing typing is an agent rather than a person. Sort by that first and the shortlist gets short fast.
Disclosure: I build PandaStack, which sits in the third category. It is not an IDE and I'll say clearly where that disqualifies it.
Three questions that pick the category for you
- Who uses the environment — a human in an editor, or a program? Humans need a terminal, an editor, port forwarding, and extensions. Programs need an API, fast start, and clean teardown. Almost nothing does both well.
- Does state survive? A workspace that keeps your uncommitted changes across a lunch break is a different engineering problem from one that is thrown away after every task. Persistence costs money and creates snowflakes; ephemerality costs setup time on every start.
- Whose infrastructure? Fully managed is fastest to adopt. Self-hosted in your own cloud is what security teams ask for the moment source code and production credentials share a machine.
The options worth shortlisting
GitHub Codespaces — the default replacement
If your code is on GitHub and your team lives in VS Code, Codespaces is the lowest-friction move. It reads the same devcontainer.json most Gitpod repos already have, prebuilds are a first-class feature, and the billing lands on an invoice you already pay. The trade-offs are the ones you'd expect from a default: you are on GitHub's machines in GitHub's regions, per-hour pricing on larger machine types adds up quickly for a full team, and stopping idle workspaces reliably is a discipline problem your finance team will eventually raise.
Coder — when the requirement is self-hosted
Coder provisions workspaces in your own infrastructure using Terraform templates, which means a workspace can be a Kubernetes pod, an EC2 instance, or a VM in your data centre, with your own networking and your own audit trail. That is the honest answer for regulated environments where code cannot leave the account. The cost is real platform work: someone owns the templates, the images, and the upgrade path, and "open a URL and start coding" becomes a project rather than a signup.
DevPod and devcontainers — when you want no platform at all
DevPod runs the devcontainer spec against a provider you choose, including your own laptop. It is client-side, open source, and unopinionated. If the value you got from Gitpod was really "the devcontainer works everywhere" rather than "someone else runs the machines", this removes the vendor without removing the workflow. There is no shared prebuild cache, no team dashboard, and no one to page when it breaks.
Daytona, Replit and the agent-adjacent tools
This cluster has moved noticeably toward AI agents over the last two years. They still serve humans, but the roadmaps are about programmatic environments: create one from an API, run a task, snapshot it, throw it away. If your Gitpod usage was mostly "spin up something disposable to run untrusted or generated code", this is the direction the market went, and a plain IDE-in-the-cloud is no longer what you want.
PandaStack — programmable environments, no IDE
PandaStack gives you a Firecracker microVM from an API call, typically in under 200 milliseconds because every create restores a baked snapshot rather than booting a kernel. You exec commands, read and write files, expose ports over a preview URL, snapshot the machine, and fork it. That covers per-branch environments, per-PR test rigs, and agent workloads where an environment is created and destroyed thousands of times a day.
What it is not: an editor. There is no browser IDE, no extension marketplace, and no settings sync. A developer who wants to open a file and type into it should use Codespaces or a local devcontainer. If you want both, the workable pattern is humans on Codespaces and automation on a sandbox API — they are different workloads, and merging them into one product is exactly why this category is confusing.
from pandastack import Sandbox
# A disposable environment per branch, created from an API call
with Sandbox.create(template="base", ttl_seconds=1800) as sb:
sb.exec("git clone --depth 1 -b feature/checkout https://github.com/acme/api /work")
sb.exec("cd /work && npm ci")
result = sb.exec("cd /work && npm test")
print(result.exit_code, result.stdout[-2000:])Start time is the feature, and it is measured in the wrong place
Every product in this space quotes a start time, and the numbers are not comparable, because they measure different things. Some measure until the container is scheduled. Some measure until the IDE renders. What you feel is the time until you can run your test suite, which includes cloning the repo, restoring a dependency cache, and starting services.
This is why prebuilds — Gitpod's genuinely good idea — matter more than raw provisioning speed. A platform that starts in three seconds and then spends four minutes on npm ci is slower than one that starts in thirty seconds from a snapshot taken after install. When you evaluate, time the same repo from click to green test run on every candidate, and ignore the marketing number entirely.
The cost model nobody plans for
Cloud dev environment bills are dominated by idle time, not usage. A developer starts a workspace at 09:00, goes to two meetings, and leaves it running until 18:00 — nine billed hours for maybe three hours of typing. Multiply by team size and it is a real line item, and every platform's answer is some flavour of idle timeout that people disable, because losing state mid-thought is worse than the cost.
So evaluate the idle path, not the price per hour. What happens after thirty minutes of no keystrokes: does the workspace stop and lose in-memory state, hibernate to disk and come back with your processes intact, or keep billing? That single behaviour moves the bill more than any hourly rate on the pricing page.
The short version
Humans on GitHub, in VS Code, who want the least work: Codespaces. Code that cannot leave your cloud: Coder. No platform, no vendor, keep the devcontainer: DevPod. Environments created by programs — CI, agents, per-PR rigs — thousands of times a day: a sandbox API such as PandaStack, Daytona, or E2B. The mistake is picking one product to cover all four, which is how teams end up paying IDE prices for machines nobody looks at.