Kristyna Zackova

Founder of ConversationLens (AI relationship coach) 🚀• Principal PM • Psych and Neuroscience Nerd 🧠• Dipping my toes in AR/VR • NYC (from Czechia)

[Ejaaz @cryptopunk7213]

sorry but this is seriously fucking impressive

china just shipped a claude code-level ai model small enough to run on your laptop.

it codes better than opus 4.5 and its tiny. 27B beating models 15X LARGER than it.

best part? shit is fully open source.

no cloud, no rate limits, no api keys

this model plus kimi k2.6 tells me open source has caught up to the frontier models

how the fuck did china pull this off?

[Kristyna Zackova @zackovak]

@RealRossRight @Replit I am not sure what issues you are running into, but I built a relatively complex app and it works perfectly. I did have issues at the beginning and assumed vibe coding just became better, but maybe it also improves with the size of your app?

[Kristyna Zackova @zackovak]

@Replit @Replit @ReplitSupport
FYI I came up with a temporary hack: I created a table in the db which contains documentation for mobile dev and asked the agent to keep it updated as it makes changes on the web. Mobile project can now read up to date api endpoints and requirements.

[Kristyna Zackova @zackovak]

My mobile development on @Replit is taking unnecessarily long. I'm missing the ability to use one codebase and db for both, mobile and web. Right now, I'm just a phone between the two apps so I can build a mobile app for the existing web app. Taking tips if anyone solved this.🧐

[Kristyna Zackova @zackovak]

Clarification: @Replit currently supports a single app type per project (web/mobile).
With an existing web app, mobile becomes a separate project and BE/FE changes must be coordinated across both.
Has anyone solved this efficiently? @Replit - is this on your roadmap?

[Kristyna Zackova @zackovak]

Some concrete examples from building https://t.co/8PMU76c7oE:
1. Internal dashboards built from masked prod data (funnels, cohorts, insights)
2. Ad-hoc agent queries like: “I just saw X happen — investigate why” (often config issues that never reach devs)

[Kristyna Zackova @zackovak]

1/6
While building https://t.co/8PMU76c7oE, I keep coming back to the same question:
Why aren’t more software companies integrating AI and vibe coding into their product workflows?

[Kristyna Zackova @zackovak]

@karpathy @dabit3 6/6
Thoughts? @karpathy, @cursor_ai, @dabit3

[Kristyna Zackova @zackovak]

5/6
I am aware of security concerns , but imo thise are solvable: read-only access, masking/anonymization, isolated envs, human review gates, guardrails, scanning.
Who is running versions of this? What tools? How are dev teams reacting?

[Kristyna Zackova @zackovak]

4/6
All of this can happen in isolated dev branches →
Then engineers harden, secure, and productionize the work.
Result: Weeks of discovery, prototyping, and early feedback collapses to 20% of time.
The time saved more than pays for engineering polish and security layers.

[Kristyna Zackova @zackovak]

3/6
Then PMs vibe code prototypes, validate ideas via conversational queries (e.g. “show churn patterns last 30 days vs benchmarks”), and auto-generate docs, tickets, and UI mocks.
This collapses weeks of work into a couple of days.

[Kristyna Zackova @zackovak]

2/6
You could give PMs read-only access to a masked subset of production data →
They do discovery directly against real-time data, market context, and experiment results → set priorities.
This alone is the most complex part of most PM workflows.

[Kristyna Zackova @zackovak]

3/3
I see this clearly in what I’m building (https://t.co/icnEiHk5Cl).
When insight becomes a trajectory instead of a moment, engagement shifts: >11% page view → signup and >60% activation, without growth hacks.
Products don’t win by being smarter than the model.
They win by creating trajectories.

[Kristyna Zackova @zackovak]

1/3
Founders building on top of AI won’t win with clever prompts or one-off insights.
If your product doesn’t compound value over time, a conversational LLM can replicate most of it in a single chat.
The moat isn’t the model.
It’s the system around it - and most teams miss why.

[Kristyna Zackova @zackovak]

2/3
The failure mode I keep seeing is session-bound value.
Users can already talk to ChatGPT anywhere - that’s not differentiation.
Behavior changes when insight persists, structure evolves, and progress is measurable over time.

[Kristyna Zackova @zackovak]

@ReplitSupport @Replit Appreciate the response. Support confirmed a shared web + mobile codebase isn’t currently possible on Replit (sent you the ticket # via DM).
This feels like a core limitation - curious if it’s on the roadmap, or if there’s a recommended workaround others use today.

[Kristyna Zackova @zackovak]

@sciencegirl The statue of victory of samothrace or its hologram

[Kristyna Zackova @zackovak]

@sciencegirl

[Kristyna Zackova @zackovak]

@sciencegirl DJ equipment

[Solomon @laz_inc]

Every company needs someone who can actually deploy AI agents. Not advise on AI strategy. Not build PowerPoint decks about the future. Actually implement agents that solve real problems across the organisation.

This skill is far rarer than it appears. Yes, all the best practices are technically available online. But you have to be paying attention to everything, from coding agents, memory architectures, filesystem and tool use, agent harnesses, context engineering, evaluation frameworks. The field moves weekly. Miss a month and you're behind.

Here's what makes this opportunity asymmetric: the knowledge compounds. Every new technique you master builds on the last. Every deployment teaches you something that makes the next one faster. The farther ahead you are today, the more prepared you are for whatever emerges tomorrow. This creates a widening gap between those who stay current and those who don't.

Companies are starting to realise they don't need a "Chief AI Officer" with an impressive title and a strategy document. They need someone with full latitude to walk through every department from sales, HR, finance, operations, to customer support and systematically identify the manual workflows, the legacy processes, the arcane workarounds that have calcified over years.

Most of these can be fixed with LLMs. Not in some hypothetical future. Right now. Today. The person who can do this consistently, who understands both the technical implementation and the organisational change management, becomes extraordinarily valuable. To any company you join, you'll look like you have a time machine. You'll see automation opportunities that everyone else walks past. You'll implement solutions in days that others wouldn't attempt for months.

This window won't last forever. As agents become easier to deploy and best practices standardise, the advantage narrows. But right now, the gap between what's possible and what most organisations are doing is enormous. If you're an high agency individual and looking for leverage, this might be the highest-ROI skill you can develop.