Hey —

This is the first Pranav’s Notes.

I’m already reading, bookmarking, and testing this stuff every week. This is where I’ll keep the parts worth passing on.

1 — Vercel is building for agents, end to end

Ship 2026 was much bigger than a normal feature roundup.

The releases now cover nearly the whole agent lifecycle:

• Agent Stack: AI SDK, AI Gateway, Workflow SDK, Sandbox, and Chat SDK.

• eve: an open-source framework for durable, sandboxed agents with approvals, subagents, and evals.

• Connect: temporary scoped credentials, so agents do not need long-lived provider tokens.

• Services, Dockerfiles, and a container registry for running more than frontend code.

• Vercel Agent: production investigation and proposed fixes, read-only by default.

My read: Vercel does not only want to host apps built by agents. It wants to be the runtime where agents build, execute, authenticate, observe, and repair software.

Read the official Ship recap:

​https://vercel.com/blog/vercel-ship-2026-recap​

2 — A model-routing claim worth testing

Martian launched Ship with a sharp promise: Opus- and GPT-5.6-Sol-like behavior at exactly half the price, backed by a model-quality SLA.

The SLA is the interesting part. It tries to cover model behavior, not just whether an endpoint stays online. But “same behavior” on someone else’s benchmarks is different from “same result” on your own tasks.

I would treat this as an eval candidate, not an automatic swap.

Announcement:

​https://x.com/withmartian/status/2079636656425165024​

Technical write-up:

​https://thesean.ai/blog/introducing-ship​

3 — Behind the scenes: the AI email client I’m building

I’m building an open-source, Gmail-native email client: ask AI to summarize a thread, search mail in plain English, or draft a reply — while you still decide what gets sent.

The work right now is mostly behind the landing page: Gmail-backed mailbox actions, drafts, review-before-send, sync behavior, and privacy/readiness details.

It is still in progress. I’m not calling it launched yet.

Follow the repo:

​https://github.com/Pranav-Bobde/better-mail​

Join the waitlist:

​https://better-mail-henna.vercel.app​

4 — Jobs worth opening

I already save a lot of current roles while job hunting. I’m going to keep a small section here, using the original application link whenever I can find one.

Observe.AI — AI Agent Engineer — Bengaluru

A client-facing role across prompts, workflows, APIs, RAG, and telephony. It asks for 3+ years and a 4 PM–1 AM IST shift.

Apply on Observe.AI:

​https://www.observe.ai/position?gh_jid=5189295008​

KeeMakr — 11 openings — Bangalore, full-time

3 senior full-stack engineers, 2 senior backend engineers, 3 product managers, 1 social media manager, and 2 video editors.

I could not find a public role page. The company-posted contact is:

sunil@keemakr.ai

Openings change quickly, so check the source before applying.

5 — One thing I published

I wrote about the part multi-agent coding demos usually skip: what happens when several agents open several PRs against a moving main branch.

It includes interactive demos for branch drift, merge queues, stacked PRs, and feature flags.

Read it:

​https://pranavb.xyz/blogs/merging-in-parallel​

That’s it for #001.

If a job is closed, a link is stale, or you know a stronger original source, reply and I’ll correct it in the next note.

— Pranav