Mohit Asthana
I carried a quota and built internal value tooling for AE teams at Salesforce. Now I build and run production AI agent systems solo, full time.
San Francisco Bay Area
What I've built
A production content pipeline, not a demo
Runs my media business end to end: a Next.js console over Supabase Postgres with a pg-boss queue handling ideas, captions, carousels, video, voiceover, newsletters, publishing, and metrics.
View the ai-marketing-os repo →A Telegram control plane, laptop off
Two bots on a small cloud droplet carry every approval and daily triage to my phone. Nothing publishes without a human tap, and the whole loop runs in the cloud.
View the human-in-the-loop doc →Cost discipline as a design constraint
Two small droplets plus metered AI usage. Cheap models by default through OpenRouter, paid renders behind an explicit approval tap, auto-generated shorts capped at 2 a day.
View the cost-controls doc →Postmortems when it breaks
Every incident gets root-caused and written up. A side app's scan, running 4 times a day, blew past the 2,000 included minutes and froze every workflow on the account.
View the postmortems repo →An AI calculator my fellow sellers adopted
Sellers picked it up on their own and used it on live deals. Built with Claude at Salesforce: 7 value drivers, a 3-year financial model, a printable one-pager.
See the tool → See the shoutout →User-facing products, shipped live
A free affordability calculator runs at caniaffordthisproperty.com, and a prompt playbook and proposal generator run my consulting sales motion. Built on Lovable, with my own domain, export, and pricing work.
See the products →Agents with real permissions, behind a real gate
Hermes runs my agents on a $12-a-month droplet. Spend and sends need an out-of-band Telegram code, code ships through a PR-only lane, and Gmail is read-only on the server itself.
See the Telegram wire →An agentic SEO system that caught its own defect
For SalesOS Labs: 300 articles drafted and critic-gated through OpenRouter, 134 live with schema markup and llms.txt. I found 100% of internal links non-canonical and fixed it, 0 to 1,213.
See the live blog → See the scorecard and one-pager →VerifiedOTE, a live SaaS we run
Co-built with a partner and live, with Stripe live-mode checkout verified in production. When a database quota took every data route down, I started the move to Postgres; the migration rehearsal runs in 34 seconds, idempotent on rerun.
See it live →Tools other people run their work on
A human editor ships every issue of a curation digest on the pipeline I built: autowriter, issue configurator, publish-ready paste for Beehiiv. Selection runs on archived performance data.
See the brand it publishes for →A money OS that caught its own stale data
An audit found every pipeline run reporting green while the bank data sat silently stale. Same day I shipped a freshness guard, outlier detection, an append-only ledger, and nightly backups.
See the dashboard →Agents write to production through a typed gate
My agents write to the core database through one MCP server, never raw SQL. Enforcement is in Postgres: a scoped role, row-level security, and an audit row for every write, including failures.
A service packaged as a plugin
The SalesOS Labs answer-engine method as a public Claude Code plugin: intake, tech audit, topic research, grounded generation, a lint gate, measurement. No client data, and no paying client yet.
View the aeo-engine repo →The system, live
The Telegram wire
Every system reports to one place: my phone, one bot each. Real messages.
Shipped products, built with Lovable
When the artifact is a user-facing product I use Lovable, then wire domain and pricing myself.
The long tail, also live right now
Everything else I keep deployed and reachable today, because a working URL beats a description.
Constraints and self-healing
MohitOS hard-caps spend and gates every paid or public action behind a human tap. A startup wizard checks its own 31-item configuration and prints the exact fix command; a watchdog restarts any process that goes quiet.
How I work
Every incident gets a writeup. Four are public in the postmortems repo, including a 12-day stall.
Cheap models handle the high-volume work. Stronger models only where output quality depends on it.
Agents generate everything. Nothing that costs money or reaches an audience ships without a human tap.
I put systems in front of real usage early and fix what the breakage actually teaches.
What colleagues say
"Mohit is a motivated self starter with a reputation for quality work and a track record of success. He is resourceful, dependable and thinks outside the box about how to solve complex problems. [...] Mohit is also organized and concise which are rare qualities that I value in my teammates."
Chaz Van de Motter, LinkedIn recommendation, public on my profile
And two straight from Slack, as they were sent.
Why forward deployed / AI ops
Carrying a quota taught me discovery, stakeholder rooms, and hunting new business instead of waiting for it. Forward deployed work is the same job pointed at a different artifact: sit with the customer, understand what they need, then build it. At Salesforce I did the discovery half; now I run the build half full time.