Agentics: the modern company won't have bullshit jobs

The modern company won’t have bullshit jobs. The upside case for AI is that this is a freeing and joyful thing.

If you or your team is trying to figure out how to use AI at your company take a look at Nori Sessions! Here’s a primer targeted towards security people. And here’s one targeted towards eng teams. And another for the ops folks. Nori Sessions is simply the best way to leverage AI tools, because we made it 😁 On average, Nori makes teams 2-5x more productive across sales, ops, and eng. Shoot me a message at amol@noriagentic.com if you are interested!

---

The year is 2022. I am the CTO of my first company. I am managing a team of brilliant engineers and a team of brilliant and completely non-technical sales and marketing folks. I spend, to a first approximation, half my life counting geese.

I am constantly in meetings. Every meeting has some complicated name, like ‘Aligning strategy for Q1’. But I know that the meeting is really about counting geese. We look at dashboards that make sure that everyone has their geese properly counted. Sometimes the engineers will mention that there were geese that they counted that aren’t reflected on the bean counter. Other times, the marketing and sales folks will say that some geese that we thought had been counted hadn’t been counted enough. There is much argument about the true number of geese.

Once we reach consensus, we spend time making sure that the goose count on GitHub matches the goose count on Linear which matches the goose count in Google Drive. There are more geese, the flow of geese is constant, so we have to spend an increasing amount of time counting them. We plan years around geese. I try to protect my team from the geese. Mostly that means that I am spending more of my time counting geese.

This is at a 14 person startup. It was worse when I was at the big tech co.

The year is 2026. I’m CEO of my second company. I haven’t thought about geese once in the last 6 months. The AI counts the geese now.

There are a lot of things to be worried about with AI. Things that I’ve written about extensively in other parts of this blog. But the thing that I am most excited about is the complete and total elimination of bullshit overhead. The future of work may be complex and crazy. But you know what it won’t involve? Having a founder staying up late at night making sure the linear story points are aligned with the things merged into GitHub so that the eng team sync meeting can save fifteen minutes of everyone’s time.

Everything that people hate, they hate because those tasks are boring and routine and dull and mindless. How many teams love to hate their CRM? Every time I get off the high of a fantastic customer pitch, I have 0 desire to then go into Hubspot or Salesforce to mark some row in a table ‘Lead’. How many hours get lost to making slide decks the right branding? Raise your hand if you lost an hour trying to figure out how to move an image that is in the background of your slide because it’s set as a ‘company wide template’, which was a feature you only just learned about ten minutes ago. How much information gets buried in slack, or docs, or linear, or wherever you do your work? Hours and hours of productive time wasted counting geese.

AI tools excel at this kind of data munging. They are so good at taking in data from calendly and piping it into salesforce or pulling in sentry data to debug something in github or whatever. Every company has a “brain.” Normally that brain is decentralized. It exists across the ten million different tools we all use to track state. Jira, confluence, notion, GitHub, linear, Asana, Monday, trello, docs, drive, slack, teams, one drive, Dropbox, one note, Salesforce, intercom, otter, fireflies, snowflake, databricks, datadog otel AWS gcloud Coca-Cola double decker bus his name was my name too.

The promise of AI is that you never need to think about any of that again. A universal AI operating system, a single entry point to the entire company brain. That doesn’t mean replacing people, though some folks may see it that way. To me it means taking the bullshit out of bullshit jobs.

I don’t think everyone is going to get there all at once, but the shape of the future is very clear to me.

• There are humans and there are agents.

• The humans spend time making decisions that matter — north star vision, strategy, marketing, and actually talking to people.

• The agents live in the background, in the ether, ephemeral, automatically bringing the vision to life with minimal friction.

This is a new and transformative way of conceptualizing work. AI is a new and transformative technology.

But it’s also very much within reach. There are teams who are “full auto” already, who are producing what would be Herculean amounts of output with a single person.

The main thing you need to get from point A to point B is a bit of forward thinking and some cloud infrastructure. We already provide the latter. Let me show you a bit of the art of the possible, in the hopes that you’ll get some of the former.

Here’s my morning a few days ago.

I wake up around 8:30AM. I walk into the kitchen and grab my phone, which I purposely try to keep in another room to avoid doomscrolling at night. There are like five slack messages waiting for me. I had a fleet of agents that were working through the night. I quickly eyeball the results while brushing my teeth.

One of them spun up at 3am to fix a bug that was reported by sentry. It posted a full root cause of the bug and a PR to fix it. I glance at the changed line diff and the pr description. Looks about right for what I’d expect for this, I happily merge it without looking at the code. I used to look at code, because I was worried about slop. But our agent hasn’t produced slop in 3 months. At some point you’ve got the thing so dialed in that it just works. I go to do some physical therapy stretches for my back — too much time coding in bed.

Shower, out the door, try to get on the Path train by 930AM.

While I’m sitting at the subway station I look through the other tasks. Two of them are confirmations to send out draft emails responding to some customer inbound. The agents personalized the responses, using a combination of call transcripts and website research. I tweak them both — I still can’t quite get the AI to nail my voice — but both emails go out in about 5 minutes. For someone who normally agonizes over each word in an email, this is a miracle.

One of my agents came back with some market research for the day. Things that the engineering hive mind on Twitter are thinking about. It is amazing for my mental health to be able to keep up with Twitter without being on Twitter. I chat with that agent a bit in slack, asking it to dig into a thread about agentjacking.

Another agent sends me a list of new papers on arxiv about representation learning. I spend the first half of the train ride reading through the summaries, asking questions about them and how they work at the stops where some 5g manages to leak down into the station.

Half the train is wearing Knicks gear, and that got me thinking that I should do something for our website. I open slack and type this prompt:

In between stops I iterate with the agent, all in slack. My cofounder sees what I’m doing and tells the bot to add a dunk animation in the same thread.

It’s done by the time I hit the 23rd stop and get out of the train. I can’t say that I have a ton of aesthetic sensibility, but I kinda like how it all came out.

By the time I get to my desk another agent has kicked up automatically, giving me an update on all of the changes in the code base yesterday alongside a brief summary of what everyone worked on. No painful daily sync, where you multiply every wasted minute by the number of people in the room. The rest of the team is locked in, talking to the agents from their CLI or slack or even lounging on the sofa on his phone. One of them is pacing around talking into his phone, recording a transcript to kick off a bunch of tasks on the go.

I am more productive on my way into work than on most full work days in 2022.

My team does 70% of our feature dev work in Slack. 30% of it from our phone. All of it through agents (we don’t write code and haven’t in months).

We do 90% of our ops work using automatic workflows that we never think about, or in slack through the agents.

We do all of our debugging through the agents through slack, though really it happens automatically before we even know there’s a bug in the first place.

We have the agents make slide decks. We have the agents update the website. We have the agents manage our CRM. We have the agents make sure linear and github are in sync.

My team gets to spend our time doing what we love. We think about product. We think about our customers. We write about things we care about, like this post. When people talk about AI being the future of work, they think that AI will get to do all the fun stuff. So far, for me, the AI has just automated away all of the things I hate doing.

And that’s really the goal, or at least should be for anyone thinking about how to deploy AI successfully.

---

Agentics is the study of how to use and reason about agents. If you are an expert in coding agents, or interested in learning more about agents, join our community slack. More articles here. Learn more about how Nori can bring your company into the glorious AI future at norisessions.com.

---