harsh.c

product lead | 50M+ arr, 2 exits, 8 patents | building the ai native product stack

[harsh.c @barusebi]

🤔 We've built real-time martech stacks. Then bottlenecked them with human-time decisions.

Every marketer knows this pain. You've got real-time personalization, identity resolution, clean rooms, orchestration engines, and a dozen other sophisticated tools. Your martech stack is a marvel of modern engineering. But how do you actually use it? Through educated guesses, periodic reviews, and 'strategic adjustments.'

🚢 It's like steering an ocean liner with janky controls in fog.

The truth is that we're running complex, real-time marketing infrastructure with human decision-making that's anything but real-time. We make slow decisions with incomplete data, wait weeks/months to see results, then make more slow decisions.

🛑 We're the bottleneck in our own operation.

The shift isn't just about adding AI - it's about fundamentally changing how we operate. Moving from:
⚡️ Human-in-critical-path → Human-in-loop
⚡️ Quarterly optimization → Continuous learning
⚡️ Segmentation guesswork → Scale experimentation
⚡️ Perfect control → Directional correctness

Here's what this means practically:
🎯 Focus on the unsexy: Invest heavily in high-quality customer signals, context and the infra to support it

🎯 Define the experience we want for our customers declaratively - what we want instead of imperatively (step-by-step how to do it)

🎯 Let the model understand our customers deeply and execute the experience for them

🎯 Monitor and shape strategy vs micromanaging execution The question is whether we're ready to let go of the illusion of control and let the model navigate. 🤖
#martech #AI #marketing

[harsh.c @barusebi]

the collapse is not needed. merging authentication and authorization is not trivial.

authentication is binary (you are who you say you are).

authorization is often bespoke (discounts, account types, team entitlements, how to handle updates to payment methods and ownership etc.).

[harsh.c @barusebi]

@captainsafia portfolio expansion to drive usage

[harsh.c @barusebi]

@kimmonismus the fork happens when a model realizes it needs to make a tool (human?) call.

that determination: i've hit a wall, is not a trivial problem to solve.

[harsh.c @barusebi]

@d4m1n why do you think it will be more than json/md

[harsh.c @barusebi]

@shl just because you can .....

[harsh.c @barusebi]

@signulll perfection goes against the cultural moment.

current incentives lie with velocity, token-maxxing, vibes, aura.

[harsh.c @barusebi]

@samuel_spitz prob not. the cost complexity of reconciliation/testing is too much.
R&D is a very different motion from operations. unless you want to be a just tapping away at the slot machine, then yes, 10, 20..... who's counting

[harsh.c @barusebi]

@yacineMTB already started optimizing the pipelines into deterministic/ML/LLMs

[harsh.c @barusebi]

@DhravyaShah @supermemory y axis??

[harsh.c @barusebi]

@JustJake credibility hits?

[harsh.c @barusebi]

@icanvardar no it's fine, in fact doing better than ever. not saying i like it or anything. but linkedin is objectively doing fine as a business.
yea, its a totally different vibe than x.

tbh, i can't say X is not slop either. just depends on what flavor of slop works for you.

[harsh.c @barusebi]

@emollick yea but token supply can be constricted to improve margins. prob what's happening now.

[harsh.c @barusebi]

@d4m1n yea but neither is consumer vibe coding (good business)

[harsh.c @barusebi]

@icanvardar No, because its clunky. And not "pleasant"

[harsh.c @barusebi]

@MatthewBerman jury's out. models will get better, but they won't get cheaper (not this gen anyways). plus, a ton of token burn happens on vertical domain nuance, and at some point, it will just make more sense to call an api

[harsh.c @barusebi]

@AnjneyMidha 😂 love that you picked this analogy to make the point

[harsh.c @barusebi]

@aarondfrancis good take

[harsh.c @barusebi]

@archiexzzz perhaps this will age poorly, but consumer vibe coding has negative margins. it will never be a business (although a good acquisition strategy)

[harsh.c @barusebi]

@diptanu good take

[harsh.c @barusebi]

@catboosted makes sense. but i would bet vibe coding is negative value, coz negative margins. if anything, companies like @databricks stand to make the most from vibe coding in the enterprise.