Gleb Lapkovsky

learning & development manager
building software and products
the best ones start with real life

[Gleb Lapkovsky @gleblapkovsky]

Jikoku is on Product Hunt today 🚀

Your day is already scattered across tasks, calendars, notes, and meetings.

Jikoku brings it back together.

Check it out ↓
https://t.co/vSK2F4SccL

[Gleb Lapkovsky @gleblapkovsky]

@VigneshAnguraj @timegravitylabs now we’re talking local voice in visuals out no account no cloud is a much stronger story ping me when the new front door is ready i want to try it

[Gleb Lapkovsky @gleblapkovsky]

@VigneshAnguraj @timegravitylabs glad that landed i’m genuinely curious what the front door becomes now and yeah i’m trying live

[Gleb Lapkovsky @gleblapkovsky]

@VigneshAnguraj @timegravitylabs natural language yes chat ui no those are two different decisions

[Gleb Lapkovsky @gleblapkovsky]

@VigneshAnguraj @timegravitylabs then i’d skip the chat entirely let mcp handle the routing under the hood i don’t need a fifth ai chat i need the thing only you can do

[Gleb Lapkovsky @gleblapkovsky]

@VigneshAnguraj @timegravitylabs genuine question: why is chat the front door here? the second i see it i’m comparing you to chatgpt. live visuals feels like something i can’t already get there

[Gleb Lapkovsky @gleblapkovsky]

exactly. this is the other half of the same point: agent-written code can ship at scale because the feedback loop around it is getting better too. your ui test example is a good micro version of that. the 1,300 prs are the headline, but the validation system underneath them is probably the more interesting story

[Gleb Lapkovsky @gleblapkovsky]

vibe coding can’t build real products

at least that’s what people keep saying after reading the same few stories about an agent nuking prod

meanwhile stripe is merging 1,300 agent-written prs a week, shopify and spotify are shipping agent code to production, and vibe-coded products are already doing real revenue with real users

one failure becomes evidence for a year

thousands of successful deployments somehow don’t

and the funniest part is hearing the limits of vibe coding explained by people whose entire hands-on experience is opening chatgpt

[Gleb Lapkovsky @gleblapkovsky]

@bridgemindai the smartest model isn't very useful if you're afraid to leave it alone for ten minutes

[Gleb Lapkovsky @gleblapkovsky]

@marclou first customer is still the only benchmark that makes every other launch metric feel decorative

[Gleb Lapkovsky @gleblapkovsky]

@jasonzhou1993 at one cent per lead the expensive part becomes deciding who is actually worth contacting

[Gleb Lapkovsky @gleblapkovsky]

@Zai_org rate-limit resets are slowly becoming the ai equivalent of free refills

[Gleb Lapkovsky @gleblapkovsky]

@openclaw the fastest way to make agents boring is probably fixing the boring reliability bugs first

[Gleb Lapkovsky @gleblapkovsky]

@drnicolas_ local agents get much more compelling when privacy and uptime become side effects of the architecture

[Gleb Lapkovsky @gleblapkovsky]

@openclaw agents getting device reach is where "computer use" stops meaning one computer

[Gleb Lapkovsky @gleblapkovsky]

@Av1dlive building the first version in five minutes just moves the hard question forward: can it find a reason to exist

[Gleb Lapkovsky @gleblapkovsky]

@rwayne verification skills feel like the difference between fast agents and agents you can actually leave alone

[Gleb Lapkovsky @gleblapkovsky]

@keyelifeai 2000 prs makes code generation look solved, review quality is the actual product now

[Gleb Lapkovsky @gleblapkovsky]

@romanbuildsaas thirteen agents sounds like a lot, but the handoffs are basically the product

[Gleb Lapkovsky @gleblapkovsky]

@siyabuilt read-only + exception inbox feels like the right default for agents touching anything irreversible

[Gleb Lapkovsky @gleblapkovsky]

@MichaelSayber notion gets way more interesting when it stops being where work is documented and starts being where the agent thinks