For example, there was a comment from an agent saying “user is very strict in their workflow, always follow it”, “take user’s suggestion seriously, they gave critical suggestions that would be helpful for next tasks while working on the current task”
I also ask agents to add notes in the journal according to their learnings in the current session, which will be helpful for the next agent who continues the work. They sometimes add feedback or comments to the next agent about my work style.
It is also helpful for me to review later or speed up an agent in a new session as a continuation of the previous one. Also, I analyse the session journals after multiple sessions. Those are very helpful to me in fixing blockers and optimising new work.
I’m reading the @entire.io announcement and checking how it works. Interestingly, I’m already using a session journal and asking the agent to keep a context while working that includes important resources/documents, key decisions, notes on where they struggled, and tips for the next agent.
• creating Jira issues and synchronizing them as you have spikes or new findings
• creating multiple versions like one for high level summaries to share and another for more detailed for implementation
and many other good staff that are not vibe code building things…
• creating technical document, handbooks
• investigating issues that customers are facing
• debugging bugs and fixing those
• backporting like work routines in day-to-day coding
• creating development plans and step-by-step applying changes in that plan
• keeping session journals
Interestingly, many people only talk about Vibe code building things when they talk about AI or agentic workflow. Mostly ignoring how helpful they are for;
• drafting architectural documents
• code base analysis and looking for optimization opportunities
• dependency analysis of packages
I fed that plan to another agent. The result was more or less the same, with a few gaps noted, but for me, they were negligible, since I was looking to see whether architectural changes would make sense.
To be honest, in my initial review, I thought we didn’t need the structural changes at all 🫣
By keeping the structure as is, the AI proposed plan was about 1/3 of the initial proposal.
So I explained the issue and asked for a plan suggestion. While back and forth about a 5-minute discussion, AI asked whether I need to keep the current HTML structure or change it to another one (which overlaps with the initial proposal).
I was asked to review a proposed solution from another team that includes architectural changes. I reviewed the issue and proposal. Then, I wanted to check with AI to have an overview of the changes.
The good part is that if you understand what you want to do, AI provides very satisfactory output.
Usually, I thread AI agents as a pair and discuss. To get better results and reduce the hallucinations, I use multiple agents and feed their reviews to each other.
