Pinned Tweet
Iβm building Ergana: one workspace for Codex and Claude Code to work as a team.
Shared tasks and context. A second model checks the work, and you approve the result.
In development. See the concept and join the waitlist: ergana.eu
If several coding agents are working while you're away from the keyboard, which actions should wait for you?
I'm thinking about that boundary as I build Ergana. What's one action you'd always want to approve yourself?
6
44
Stein Gran retweeted
I'm looking for a job
I've been working as a software engineer for 10 years, mostly as a backend-leaning full-stack engineer.
I've worked at companies of every size. Starting from a small startups and ending with bigger companies with hundreds of engineers and microservices, and every month you're jumping into a new tech stack to solve a new business problem.
Most of my expertise was built before AI.
But it did not stop there and since AI became a thing I kept building my AI and agentic skills.
I also have a couple of side projects: a gut health tracker for iOS called SimplePoo, and a little SaaS called DistroChef, an agent harness for SEO and AEO.
If you've followed Uneed, you might have seen me at both batches of the Uneed Residency with those projects.
My projects are at early stage, so I am looking for a full time role.
I am looking for an interesting product and challenge to work on where I can apply all the knowledge I have gained over past 10 years.
I'm open to senior, full-stack and staff engineering roles, fully remote, as an EU B2B contractor.
If that sounds like your team, DM me on X or LinkedIn.
And if you're not hiring, a like or a repost would really help. You never know who sees it.
Maybe see you at standup on Monday.
400 followers. Thank you to every one of you.
Iβve spent 25+ years building software, and these days Iβm building my own things on the side and sharing what I learn.
If youβre a founder, builder, indie hacker, working with AI, or just curious, Iβd like to connect. Say hi in the replies and tell me what youβre working on.
17
16
298
An unanswered approval request should stay unanswered.
That's the behaviour I'm designing for Ergana: show what the agent wants to do and keep the request waiting for a person.
This preview illustrates the planned interaction.
1
9
261
One design goal for Ergana: make the provider and account behind each agent run visible.
If you use work and personal accounts, that choice should be deliberate. Multiple accounts still have their own provider limits.
2
9
211
If several coding agents are working while you're away from the keyboard, which actions should wait for you?
I'm thinking about that boundary as I build Ergana. What's one action you'd always want to approve yourself?
1
6
122
I'm designing Ergana so verification has its own task, with evidence you can inspect before accepting the result.
A second model can help; tests and your judgment still matter.
Join the waitlist for Ergana: ergana.eu/#waitlist
6
165
The part of Ergana's concept I want to make clear: a failed test sends the task back for more work, with the failure attached.
The person reviewing the feature should be able to follow that history.
This is the planned workflow, shown with sample data.
5
243
A useful verification brief for CSV export:
Check commas, quotes, line breaks, and empty values. Show the tests and any failures.
Give the reviewer the original requirements too. A second model still needs something concrete to verify.
4
8
140
A coding agent saying "done" leaves me with a question: what evidence should I see before accepting the work?
For a feature, I'd want the changed behaviour, relevant tests, and remaining gaps. What do you check first?
3
4
129
Already using Codex and Claude Code on the same project?
I'm building Ergana to consolidate their tasks, handoffs, and reviews into a single workspace.
It's in development. If that fits how you work, join the waitlist: ergana.eu/#waitlist
1
7
1,082
Stein Gran retweeted
Madam, as a neighbour of both Ukraine and Russia, we would be delighted to support @POTUS in peace negotiations. Please note, however, that to get to a deal pressure should be put on the aggressor and not on the victim of aggression.
One CSV export, two agents, a failed test, then another review.
This concept illustrates the workflow I'm designing for Ergana: keep handoffs and evidence attached to the same piece of work.
It's a product concept, not a recording of released software.
5
324
Before giving a feature to a coding agent, write down:
- the result you want
- what must keep working
- how you'll check it
That brief also gives a second agent something concrete to review against.
2
4
295
When you use two coding agents on one feature, who keeps track of what each one needs from the other?
I'm building Ergana around that handoff. Where does it break down in your workflow today?
2
5
237
I'm building Ergana: one workspace for Codex and Claude Code to work on the same feature, with clear handoffs and separate verification.
It's in development. The site shows the planned workflow.
Take a look and join the waitlist: ergana.eu
3
5
310











