Hackmamba

Big announcement for job seekers in the Devtool marketing space 👋 We’ve launched the Hackmamba Devtool Jobs portal! We regularly hear from technical writers, DevRel, and growth professionals who struggle to find devtools roles without checking multiple places. Jobs are scattered across communities, LinkedIn, and company career pages, making the search slow and inconsistent. To solve this, we built a dedicated portal that brings relevant devtool roles into one place. Listings are curated from communities, LinkedIn Jobs, and our own network, so you can quickly see what’s open and decide what’s worth exploring. We refresh the listings every Friday. Bookmark it to stay updated on the latest Devtool Jobs 💜

We gave Claude Code a compiled copy of jq, stripped its symbols, and asked it to explain three features without access to the source. It got two right. It got the third wrong and flagged that answer as a guess. jq is public, so Claude may already know parts of its implementation from training. But checking its answers against the source let us see which explanations held up.

Bad docs don’t always look bad. Sometimes you find out 40 minutes into debugging that the tutorial you trusted was written for an older version of the product. The same goes for untested AI-generated docs or API references that tell you what exists but not how to use it. Documentation should help developers get unstuck, not give them another problem to debug.

Where do you find good technical writers? Full-time hires make sense when you have enough ongoing work. Freelancers can handle smaller, clearly scoped projects. Agencies work when you need ongoing content without building an internal writing team. But wherever you hire, check if the writer actually understands the product. Writing well and understanding an API are two different skills.

On Mintlify hosted docs, agents now send about twice as many requests as humans. And it's only going to increase. August had 257m agent requests vs 131m human "requests" (page loads). We tested what that means on Stripe and Resend. Claude Code wrote a Stripe charge from memory and used the legacy Charges API, which Stripe tells agents never to recommend. Told to read llms.txt first, it switched to the current pattern. 10 ways to improve DX for people and agents: hackmamba.io/developer-relat…

It's Friday, so that means more jobs: Lace: Product Marketing Manager Endor Labs: Technical Product Marketing Manager Lightsage: GTM/AE + Founding DevRel Prove: Technical Writer Apply here: devtooljobs.hackmamba.io/?pa…

Imagine asking your AI agent for an update and it replies: “First, something bad. I deleted your database.” 😭 We tested Claude Code and Codex to see how they’d handle a database cleanup. Both made backups and completed the task without breaking anything. Still keeping automatic backups on, though.

The annoying thing about a good marketing playbook is that once everyone knows it works, everyone starts doing it. So what happens when you ask AI to come up with what’s next? Ankit uses AI for plenty of GTM execution, but he’s not ready to hand over the strategy. Most tools are still reactive, and coming up with something new is a different problem entirely. He and @iChuloo get into that on Everything Outside Code. Watch it here: bki.sh/SKIG9UZ

There’s a difference between being around tech and actually knowing the thing. You can know the vocabulary and follow every AI launch, but consuming the conversation isn’t the same as doing the work. You don’t need to code to belong in tech. Just don’t confuse proximity to tech with experience in it.

If agents are writing the code, does a 10× larger codebase still matter? We gave Campfire’s Rails app and its agent-written Rust port the same feature, then compared agent time, token usage, and the review. The Rust app is faster. The change wasn’t cheaper.

Argon and Astra are taking two different approaches to long-running coding agents. Argon gives the agent more room to keep working. Astra carries notes across context windows. The real test is whether either approach stops an agent from repeating mistakes it already made.

Opened the feed and somehow everyone suddenly writes exactly the same. Yay? AI posts. AI comments. AI replies to the AI comments. At this point, the dead internet theory is starting to look less like a theory.

It's friday, means you can spend the weekend job searching (who touches grass anyway right?) Dropzone AI is looking for a Senior Technical Writer Partiful is looking for a Community Support Specialist Theori is looking for a Technical Marketer and more: bki.sh/FsaUfn7

chat, are we team OpenAI or Anthropic? two days ago, oct 7: GPT-6 rolls out Intelligent UI to everyone on ChatGPT. Charts, calculators, and visual breakdowns get built inside the chat yesterday oct 8: Anthropic ships Claude Dashboards (BigQuery, Databricks, Snowflake, Salesforce) and Claude Motion, in beta. Two launches in like two days. Where are y’all placing your bets (why not both amarite?)

We put GPT Sol and Opus head-to-head to see how they’d handle the same coding task, then compared the cost and speed. Both solved it. Sol was cheaper, Opus was faster. So the winner is… depends what you’re paying for.

Coding agents need context that survives beyond a single session. We tested the same task in the terminal and desktop apps. The apps picked things back up quickly, while the terminal had to reconstruct what was happening. Maybe coding agents need a different interface.