I've done the same but using Notion, and I've found really interesting to let the agents write a small "Dev Blog" post on a Notion page with what they built in the sessions, with screenshots, videos etc.
Also working with parallel sandboxes is a must: I use a manger session that read/write the issues from Notion and dispatch them to workers, it handles similar work to the same box to minimize merge conflicts.
PS: I'm using AgentBox for this (discl: I'm the author, free OSS): npm -g i @madarco/agentbox but now also Claude is releasing Projects (still rolling out) and Cursor had cloud agents for a while
nono is great but you'll still won't be able to run multiple dev servers, dbs or test in the browser...
the better alternative is using sandboxes, for TRUE isolation.
but most people skip them because it's harder to set up a dev environment in a sandbox, for this I built agentbox [1], a tool to teleport your files, skills, settings, claude/codex subscriptions etc in any of the popular sandboxes. eg:
agentbox e2b claude
or:
agentbox hetzner codex
each session is persistent, with a full VM, with a browser, webVNC screen sharing and local https tunnel for easy preview on *.localhost domains.
Also git credentials never leave your pc: when a box needs access it passes through the agentbox service running on your machine, which allows PR creations, comments but only on the box branch, no deletion etc.
> nono is great but you'll still won't be able to run multiple dev servers, dbs or test in the browser...
With our profile setup you can do all three of those things - there is a fresh chrome profile that is blocked against the same network firewall. Claude and Playwright can navigate to the dev stack and screenshot/test it (and non-allowlisted websites are blocked at the network level).
As long as the developer runs the `task start` command outside the sandbox (because docker is blocked), the agent can then interact with it all that they want, and our devs can run 2-3 of these per machine and work on concurrent projects. (based on memory usage / cpu usage)
> a tool to teleport your files, skills, settings, claude/codex subscriptions etc in any of the popular sandboxes
FWIW this is _explicitly_ the opposite of our threat model - everything other than your git credentials that you 'teleport' in (code, settings, Claude/Codex subscription tokens) now lives inside the blast radius, on third-party infra, and a prompt-injected agent in the box can send it anywhere unless you're also egress-filtering. A VM isolates the machine - it doesn't protect the data.
Much easier to keep it all local and refuse any network or file access to anything you don't explicitly approve
It had already been like this long before widespread LLM adoption: quality hiring was only really possible through manual candidate scouting on LinkedIn, at conferences, through word of mouth, and so on.
Sending a CV had already become mostly useless 3–4 years ago because of the huge amount of noise: candidates applying from all over the world, often even spoofing their actual location, FAANGs firing, flooding the market with (in theory) great candidates with great resumes.
The solution is the same one that has been successfully used forever: a trial period.
Luckily, a video interview with a senior developer is still enough to spot a good candidate.
Go through real code: add a bug to a branch of your codebase, have the candidate share their screen on TeamViewer, and let them debug and fix the issue. Ask questions live to understand how they reason about the system, how they would test whether the change works, and so on.
This will filter out 99% of candidates. But it is still possible to get lucky, which is why the trial period matters.
I’ve never had major issues and have always hired very strong engineers. I only had to terminate someone after the trial period twice.
In US it is at-will employment. What would be the difference really between a trial period and just letting an employee go 2 months or 4 months of 7.83 months down the road. I mean, in at-will employment trial period sounds just like an artificial gimmick that the employer forces onto the employee's mental state.
At-will (or not) employment is only relevant in the absence of a contract. Most/all tech employment has contracts with terms of dismissal. And in my experience this is generally a good thing for everybody. I've lived/worked in countries with for-cause employment and it's overwhelmingly more common for it to be used for abuse than protection - somebody gets fired for very good reason, but then sues just because it's generally cheaper/quicker/better PR to settle than litigate. Montana probably has the best idea of a global probation - 12 months at will, and then the employee can only be dismissed for cause after that.
It is a great model, but many HR departments will not let you do this. Every termination becomes an exercise in endless documentation, including what you thought was an open-and-shut, by the numbers trial period.
In the UK, employment law is strongly on the side of the employee, and so I'd understand HR departments trying to do everything they can to protect the employer. But having a (typically 3-month) probation period is common in almost every company, and it's not uncommon for people to be let go in the third month when they haven't been up to standard. Everybody knows it happens, and in fact you can see people's attitudes shift - for the first 3 months, every new employee is tiptoeing around very gentle, trying not to offend anyone, going above and beyond so that they are seen to be making a positive impact. After the probation period ends, most people revert to a "normal" working attitude, everything's a bit more relaxed, you'd push back on unreasonable demands, etc.
In somewhere like the US, where the laws strongly favour the employers, I'm surprised that HR departments make it so hard to fire people in a trial period. If you can fire people for any reason, underperformance in a trial period seems the most safe reason possible for dismissal.
The opposite extreme is France where employment law favours employees so much, many companies are reticent to hire at all because it's so hard to fire someone once hired.
I don't know. To me it feels like a dying genre. If some areas can keep it alive that is great, but I only need to look at local areas - arcade games and also areas went almost completely extinct here.
I do the same, and faced the issue that claude/codex loose context when doing subtasks (and subagent don't have plan mode).
So I've built Agentbox to be able to launch from claude/codex multiple VMs with claude/codex (can also mix).
The parent agent watch for prompts and questions, enforce /review, /simplify, that the sub agents file a PR and wait for bugbot comments etc.
This way the parent agent running in a /goal don't loose context, enforce a good workflow, manage the backlog and parallelize/merge back the work on the main repo
you are right, I've experimented with cp -a on macOS as well for https://github.com/madarco/agentbox and in the end found it's actually faster to use worktrees inside docker containers while mounting your .git repo inside them.
Then after the node_modules (or apt packages) are installed, take a docker commit snapshot.
Now I have truly isolated parallel workspaces in <10s.
Also the system was easy to adapt to cloud environments as well so now I have Hetzner, Vercel, Daytona as well (using their native snapshotting systems for fast boot after the initial setup)
can't agree more, I now run my agents in parallel with "agentbox claude", "agentbox opencode" and it teleport my project and settings to an hetzner VPS
That's what your nose is for. (I'm quite skilled at advancing or going back by gently tapping the kindle against my face. It helps that I'm very nearsighted so it's kind of already there)
Only way to get buttons that are comfy on both sides would be to flip it screen out. That doesn't seem like a very useful position to hold a reader in though.
If you turn it upside down the left hand buttons will be way too high on the reader
This looks great as well! :) Love the addition of a creative element. I also recently launched a new one: https://spaceword.org (90% of the code is written by cursor)
Also working with parallel sandboxes is a must: I use a manger session that read/write the issues from Notion and dispatch them to workers, it handles similar work to the same box to minimize merge conflicts.
PS: I'm using AgentBox for this (discl: I'm the author, free OSS): npm -g i @madarco/agentbox but now also Claude is releasing Projects (still rolling out) and Cursor had cloud agents for a while
reply