|
| PRANAV'S WEEKLY COMMIT |
#001 · JUL 2026 |
LEAD STORY · VERCEL SHIP
Vercel's agent stack—and six more things worth opening.
AI, building, writing + jobs. Weekly-ish.
|
|
Hey —
No big intro — just a few interesting things from recent weeks.
Some are launches I kept thinking about, some are things I’m building or writing, and a few are job openings that looked worth saving. These are the ones that stuck.
|
|
01 · THE BIG ONE
Vercel Ship, in three layers
The part that caught me wasn’t any one announcement. It was how neatly the pieces now line up across the whole agent lifecycle.
The easiest way I found to make sense of the release is in three layers:
Build agents — AI SDK, Gateway, Workflow, Sandbox, Chat, Connect, and eve. |
Run apps + services — Dockerfiles, Vercel Container Registry, backends, queues, cron, MCP servers, databases, and Vercel Services. |
Operate in enterprise — Passport, Security Dashboard, BYOC, and Vercel Agent. |
My read: put together, this feels less like a feature roundup and more like Vercel making a bid to become the default runtime for agent-built software—where agents build, execute, authenticate, deploy, observe, and even repair software. That’s the part I think is worth watching.
See the full Ship catalog →
|
|
02 · WORTH TESTING
Martian's 50%-cost claim
Then came Martian’s Ship launch with one of the sharper claims I saw: Opus- and GPT-5.6-Sol-like behavior at exactly half the price.
|
HALF THE PRICE.
While claiming the capability + behavior of the reference model.
|
What makes this more interesting than the usual “cheaper model” announcement is what they say they’re preserving:
|
Capability — solve the same tasks. |
|
Behavior — preserve tool use, formatting, refusals, and instruction following. |
|
SLA — make model quality, not only uptime, contractual. |
My take: I wouldn’t swap anything blindly on this claim alone. But it is exactly the kind of thing worth running through your own evals—same prompts, same tools, same failure cases. Their benchmarks tell you where to start, not whether it works for your workload.
Read the technical case → · Announcement
|
|
03 · BEHIND THE SCENES
The AI email client I'm building
Meanwhile, most of my own build time has been going into Better Mail—an open-source, Gmail-native email client for summarizing threads, searching in plain English, and drafting replies while you still decide what gets sent.
Where it is right now: I’m in the less glamorous, but important, part—mailbox actions, drafts, review-before-send, sync behavior, and privacy details. It is still in progress, but it is starting to feel more like a real product than a landing page.
Follow the repo → · Join the waitlist
|
|
04 · FRESH OPPORTUNITIES
Seven opportunities worth opening
Since I’m actively job hunting too, I keep running into roles that look genuinely worth opening. I’m starting to collect a small batch here in every issue—using direct job links where possible, and the original X hiring post when that is the only source.
|
Observe.AI — AI Agent Engineer
Bengaluru · 3+ years · 4 PM–1 AM IST
Model APIs, RAG, vector databases, telephony, and client-facing delivery. Apply →
|
|
AnswerThis — Product Engineer
Remote India · ₹20–35L
Own features end to end across Next.js, FastAPI, and AI/RAG. Apply →
|
|
HighLevel — Full Stack Builder (Team of One)
Remote India · AI & Marketplace
Own AI-native products from idea to production; React/Vue/Angular + Node/Python/Java. Apply →
|
|
SideShift — Software Engineers
New York City · $130K–$200K + equity
Part of a broader NYC hiring round across engineering, sales, and operations. Hiring post →
|
|
Tenex — Applied AI, Engineer & Creator
New York City · $180K base + bonus
Build production AI and teach through content, workshops, and live training. Apply →
|
|
Aside — MTS · Founding Designer
San Francisco · $150K–$230K + equity
Design the browser as a form factor for personal AI; relocation support included. Apply →
|
|
KeeMakr — 11 openings
Bangalore · Full-time · Engineering, product + content
3 senior full-stack, 2 senior backend, 3 product manager, 1 social media, and 2 video editor openings. Hiring post →
|
I saved these recently, but openings move fast. If one has already closed—or you find a better original link—reply and I’ll fix it for the next commit.
|
|
05 · TWO THINGS I PUBLISHED
Two rabbit holes from recent weeks
I also published two pieces that started as questions I kept coming back to:
|
AI COST · 11 MIN
Why lower token prices don't lower AI bills
This started with a mismatch I kept noticing: models get cheaper, but real AI workflows do not always follow. The more useful unit is cost per finished task—after retries, context growth, tool calls, and failures. Read →
|
|
MULTI-AGENT GIT · 4 MIN
Multiple AI agents, multiple PRs—how does this not break?
This came from a more practical worry: once several agents open several PRs against a moving main branch, how does that not turn into chaos? Worktrees isolate coding, not integration. The interactive demos cover branch drift, merge queues, stacked PRs, and feature flags. Read →
|
|
|
|
| Commit #001 complete. |
| Broken link or better source? Reply and I’ll patch it. |
|
|
|
|
|