|
Sponsored
|
🛠️ worktrunk Makes Git Worktrees Easy
⭐ 10.4k · max-sixty/worktrunkRepo
Running more than one coding agent at a time sounds efficient until you're the one typing git worktree add -b feat ../repo.feat for the fourth time today, then cd-ing into it by hand because muscle memory still thinks you're on one branch. That's the tax nobody budgets for when they start fanning out agents across a codebase. Today's hero fixes exactly that friction, and one of the drops below is for anyone who's hit the other wall: the one where your agent remembers nothing between sessions.
A CLI that makes git worktrees as easy to create and switch between as branches.
Running 5-10 agents in parallel means a fresh worktree for each one, and native git worktree commands force you to type the branch name three separate times just to get a new directory open.
Quick start: brew install worktrunk && wt config shell install
Catch: it's built specifically around parallel agent workflows, so if you're on one branch at a time the value drops fast.
HubSpotAffiliate
HubSpot, a CRM with prebuilt AI agents that run on the customer record. See how HubSpot works We may earn a commission.
mem0Repo ⭐ 68.3k
Drop-in memory for AI agents so context survives past one session, no rebuild needed. Catch: the benchmark numbers quoted are from their managed platform, not the open-source SDK, so expect similar direction but not identical gains.
agent-desktopRepo ⭐ 12.3k
A CLI that lets an agent see and operate desktop apps without screenshots or pixel matching. Catch: snapshots and clipboard reads can carry private desktop content, so keep them out of logs you share.
stack-diagramMCP ⭐ 1.5k
An agent skill that turns a one-sentence stack description into an architecture diagram with official logos. Catch: it's a documentation tool, not an infra one.
|
From Our Partners
|
New here? Start with The Awakening. It's free, it takes one sitting, and it's where every reader should begin.
Turn a messy project debrief into a review meeting nobody can dodge
Start Here is one prompt from our pack, every issue, tested on ChatGPT and Claude. Copy it, paste it into the free app you already use, and fill in the four bits in capitals: what the team just finished, how big the team is and what it does, and where the tension already sits.
The team: [TEAM SIZE] people, [TEAM CONTEXT]
Where the friction already is: [KNOWN TENSIONS]
Design a 60 minute agenda that gets past the surface, where everyone says the same three polite things and nothing changes.
Give me, in this order:
1. The opening question, one sentence, to replace what went well. It has to be answerable by the quietest person in the room.
2. Two or three exercises that pull out the problems built into how this team works, rather than one off complaints. For each one: how long it takes, what people actually do, and how it runs with [TEAM SIZE] people.
3. How the group picks what to fix, given it cannot fix everything. Name the method and say who decides.
4. The close: how commitments get written down so they are still alive the next time you do this. Every commitment needs a name and a date.
5. The moment this meeting is most likely to go wrong, given [TEAM CONTEXT] and [KNOWN TENSIONS], and the sentence I use to pull it back.
Rules:
- Use only what I told you about this team. Do not invent an incident, a person or a falling out.
- Nothing in the agenda asks anyone to criticise a named colleague in front of the group.
- If what I wrote under Where the friction already is says there are none, design the meeting to find them rather than assuming there are none to find.
- Give the timings as minutes that add up to 60, and show the total.
The line that makes it work: "Design a 60 minute agenda that gets past the surface, where everyone says the same three polite things and nothing changes." Most review prompts ask for a nicer version of the standard retro template. This one tells the model the default output is the failure state, which forces it to design around your specific team instead of reciting a format.
TRY THIS: open ChatGPT or Claude right now and fill in just the first blank, what finished, with your last shipped project. Run it once before you add the harder details. You'll see the gap between a generic agenda and one built for your actual friction.
This is one of 100 in the Aigent prompt pack. The other 99, plus guided builds with Mike on video, are inside the Aigent membership: Join the Aigent membership. Seven days free, then $29 a month. One click to stop.
|
EARN FROM HOME
Get paid to train AIAffiliate AI companies pay people to rate, fix and check what their models write. Open roles in the US list typical pay of $35 to $150 an hour, fully remote, on your own hours. Browse live roles and apply free. We earn a referral fee if you're hired through this link. It never changes your pay. |
A paper found frontier models can interpret and modify raw model weights. Simon Willison shows a 27B Qwen model on his laptop can at least read them, correctly naming the game a policy network was trained on. Weight-level interpretability just got a lot cheaper to try. (Simon Willison)
A UEFI bootkit is infecting fully patched Windows machines. BlackX bypasses Secure Boot on fully patched systems through the old Baton Drop flaw, and the fix only bites if an admin manually triggers the certificate revocation, which Windows updates don't do on their own. If you're deploying agents on Windows infrastructure, patched is not the same as safe. (Ars Technica)
Six in ten US shoppers who used AI for a first-time purchase rejected a product on its advice. Attest surveyed 4,000 consumers and found 62% of US AI-assisted first-time buyers ruled out at least one product on the AI's say-so without checking another source. If you're building anything that recommends rather than retrieves, that trust cuts both ways. (Attest via Retail Dive)
Vercel CI now runs on a private GitHub repo without a linked Vercel project. Until now you had to create an empty project just to connect the repository; enable it with vercel ci enable from a clone. Small fix, but it removes a setup step that tripped up anyone wiring CI into a repo that wasn't a Vercel deployment. (Vercel)
|
Recommended reading
If you like The Aigent, a small group of operator-tier publications worth your inbox: see the shortlist. |
Which section do you open this issue for?
Reply and tell me what's eating your afternoons right now: the worktree juggling, the agent that forgets what you told it yesterday, or something else entirely. The repos in Drops only earn their spot if they're solving a real problem you have, not a hypothetical one.
See you Thursday.
Haven't taken The Awakening yet? It's free and it's where we'd start. Take The Awakening →
|
Want to reach builders shipping with AI every weekday? Advertise with The Aigent. The Aigent stays free for everyone, and reader support keeps it that way: Support The Aigent. |
