Back

Blog

Yeh tokens mujhe de de thakur

Do not want to be one of those. But something changed inside of me the first time I put Codex on a goal, went to sleep, and woke up to work that was done, work that would have taken me a few days to finish.

Now, I'm not an AGI doomer and I never will be. I want to write down how I think about the future of work. B2B is the only world I understand, so let's just talk about that. We've always taken human labour and converted it into capital. Now that all of that is being absorbed into tokens, not that much changes. There's always going to be prioritisation of work.

Assuming every company only has so many tokens to expend (just like they had only so many human labourers to expend, of course they could hire more, just like they can get more tokens, but that comes with a cost, not just monetary but also from a POV of bloat, managerial overhead, etc.)

Now, why would a company choose to expend those tokens on non-crucial workloads, anything that is not advancing their business, especially because those very tokens are being used by their competitors to advance every single day? It’d be criminal for them not to take every single token they can get and point it towards the advancement of their biz. So hypothetically, even if they do get more tokens, because a lot of their barriers, which were barriers of humans being humans, are now gone, those should be spent towards acquiring more customers, servicing their customers, expanding their product and business, etc.

Not at rebuilding things they could be using vendors for. Not expend tokens at something they can borrow for from vendors. Every potential usage of tokens will be prioritised, and everything that is not high on the list should be borrowed tokens, not used ones.

As for vendors, there is now a lot of value in choosing which problems to work on. Optimise how good your token is at achieving the same results than if your customers did it themselves. I suppose I also mean be more token-efficient for higher-quality output. And this also means producing reusable / recyclable work, so that your cost of tokens is spread across more customers and services. The ROI of each token spent should be maximised.

Finally, vendors should also aim to maximise the LTV of their token's output for more return on investment.

The more I think about it, none of this is different from the world pre-AI. It's just a lot, lot, lot more accelerated. I could replace tokens with humans and it would alll still make sense so I think the kids are going to be okay.

Why building on SORs sucks

Dearest gentle readers, I’ve had basically the most intense 4 weeks of my life and feel compelled to write this because I find myself suddenly having strong opinions about everything. I guess YC does that to you.

Coming back to what’s up with the biz - Let’s start with a little background. We started Ressl AI in the first half of 2025 - with the thesis that productising IT services will be the next big unlock in capturing enterprise spend as models get better and better at coding tasks. We chose Salesforce as our first platform of attack for a multitude of reasons. > Salesforce was a front-office software. People seemed to care about the speed of delivery of configuration changes in front-office software more than back-office software. We also believed that one could make a revenue play here. “If your Salesforce implementation sucks, your pipeline leaks X% of deals leading to Y impact in business”.

> Salesforce as a PaaS was the easiest to connect to (mind you - this is pre-MCP popularisation - yes it hasn’t been that long). General purpose coding agents still sucked for Salesforce tasks in Q1 2025, but we had hunches of how we could get them to work for Salesforce to start off with - until the technology improves enough to use our distribution in Salesforce and go cross-platform.

> Least important reason but unlike SAP etc - Salesforce has mid-market customers - who I could build with as design partners before I needed to go enterprise. And, a senior of mine from college ran a Salesforce consulting firm and he let us work for him for a few months.

We launched in July after being a services firm for a few months. The next few months were a grind. We did well considering the space. In every offline and online forum, we went to we were known as the config agent for SFDC. We had ~400 tickets being solved by the platform on a weekly basis at our highest. I was known as the Salesforce girl, to my dismay. I can now say I did not enjoy being @-ed at every time someone mentioned Salesforce.

Now, here are a list of things I have learnt in few months of being in production that might be useful if you are exploring this space:

Now, I must admit I do not know how the configuration of the rest of these platforms work - but I understand Salesforce deeply and anything worth configuring is totally possible API-first which means unfortunately for me, Cursor/Devin/Claude Code for Salesforce is just Cursor/Devin/Claude Code.

(AT THE 2025 OPENING OF DREAMFORCE BY FUCKING MARC BENIOFF - THE FIRST THING HE ANNOUNCED TO THE CROWD WAS AGENTFORCE VIBES. Can you imagine what it felt like to be building what I was building and sitting in the crowd listening to that keynote

Last point I’ll add here is about procurement. Things are crazy in AI and procurement teams are literally losing their minds. You need to be very clear with the massive ROI that you’re bringing to the company. It can not be a slight uplift over an existing product.

Some somewhat bull points:

What’s next for Ressl?

Original post