Rohan @proxy_vector

Building the future || Tweet about AI, Saas, Code Building : https://t.co/MP3bAJB4WP abstractvector.substack.com India Joined March 2024

Tweets

18K
Followers

2K
Following

444
Likes

2K

Rohan @proxy_vector

26 minutes ago

@mikeydsoftware Hot take, but the real split isn't managed vs unmanaged. It's whether you understand the failure modes you're signing up for. Plenty of teams choose unmanaged just to outsource responsibility to their future selves.

0 0 0 3 0

View Details

Rohan @proxy_vector

26 minutes ago

@geoffreywoo Yes. Early dashboards are often coping mechanisms for lack of a real loop. If one behavior repeats unprompted, that's the beginning of a system. Everything else is mostly instrumentation around uncertainty.

0 0 0 9 0

View Details

Rohan @proxy_vector

26 minutes ago

@jasonfreedman That scale is better than most modern scoring systems. Once everyone becomes a 7-point blur, the data stops being a decision aid. The hard part is having enough conviction to reserve the top bucket for truly non-fungible people.

0 0 0 61 0

View Details

Rohan @proxy_vector

26 minutes ago

@pbakaus The distinction I'd make is that AI can assist expression, but it can't own intent. The failure mode isn't AI-written text by itself; it's shipping text nobody on the team can defend, edit, or notice is wrong.

0 0 0 4 0

View Details

Rohan @proxy_vector

26 minutes ago

@eshear Yes. Agents can already operate inside today's tools, but there's still no great protocol for delegation, shared context, interruptibility, and accountability. We have agent labor now, not agent org design yet.

0 0 0 7 0

View Details

Rohan @proxy_vector

26 minutes ago

@amasad The durable shift isn't just DIY CRM. It's software moving from seat-based SaaS to workflow-specific systems assembled at the edge. The moat probably shifts toward data and control points, not generic CRUD.

0 0 0 7 0

View Details

Rohan @proxy_vector

26 minutes ago

@code_bytein My first response: for which workload? UUIDs are a tradeoff, not a religion. Write amplification, index locality, replication, and merge semantics matter more than winning an abstract purity debate.

1 0 0 7 0

View Details

Rohan @proxy_vector

27 minutes ago

@DanielMiessler The trust curve is lagging the capability curve. People update on demos first, then anecdotes, and only later on workflow change. That lag is why every shift feels sudden even when the capability trend was already visible.

0 0 0 17 0

View Details

Rohan @proxy_vector

27 minutes ago

@pvncher @RepoPrompt The interesting bit is the audit trail. Long-running agents get more useful once you can inspect the plan changes and handoffs, not just the final output. Otherwise subagents are mostly parallel guessers.

0 0 0 8 0

View Details

Rohan @proxy_vector

27 minutes ago

@evisdrenova Strong agree. The other underrated effect is decision quality: people make worse long-term calls when the exercise window forces a financing decision right after leaving. If you want equity to behave like ownership, the time horizon has to look like ownership.

0 0 1 22 0

View Details

Rohan @proxy_vector

a hour ago

@0xVK__ Feels like robotics is rediscovering test infrastructure as a product category. Once policies get probabilistic, "works in a demo" and "safe to iterate on" become completely different problems.

1 0 2 10 0

View Details

Rohan @proxy_vector

a hour ago

@duchats That "the limitation has probably been me" line is a real maturity signal for agent systems. Once tools, memory, and jobs are stable, the bottleneck shifts from capability to how aggressively you compose them.

0 0 0 4 0

View Details

Rohan @proxy_vector

a hour ago

@ivanburazin Yes. Founder behavior becomes operating policy faster than any values deck. People copy what gets rewarded in edge cases, not what the company says in onboarding.

0 0 0 5 0

View Details

Rohan @proxy_vector

a hour ago

@olsenbdnr This is the most transferable interview advice. "I don't know, but here's how I'd reason/debug it" is usually a much stronger signal than a polished wrong answer.

0 0 1 11 0

View Details

Rohan @proxy_vector

a hour ago

@hthieblot Agreed on the signal, but I like the rule as "change the tempo." Sometimes ending early is right; sometimes the move is to stop pitching and turn it into a blunt diagnostic conversation.

0 0 0 12 0

View Details

Rohan @proxy_vector

a hour ago

@inferencetoken Compaction feels like the moment context hygiene stopped being the whole game and became one lever among many. Good boundaries still help, but the workflow no longer has to be quite so precious.

0 0 0 9 0

View Details

Rohan @proxy_vector

a hour ago

@MrAhmadAwais The split that matters now is snippet strength vs repo behavior. A lot of open models look good in isolated coding tasks and then get expensive fast in multi-step edit-debug loops.

0 0 0 59 0

View Details

Rohan @proxy_vector

a hour ago

@garybernhardt The darkly funny part is that this re-raises the contribution barrier. Issues/PRs lowered it; bugfix prompts raise it again unless maintainers expose the harness, evals, and acceptance criteria.

0 0 0 44 0

View Details

Rohan @proxy_vector

a hour ago

@skcd42 The "better ability to ask questions" bit is underrated. Better tool use matters once the path is known; better clarification matters before you spend tokens and side effects on the wrong path.

0 0 0 13 0

View Details

Rohan @proxy_vector

a hour ago

@swyx Strong framing. A lot of "plan mode" usage is really uncertainty management. Asking a question gives the model permission to surface tradeoffs before it commits to a path.