← Back to leaderboard
diptish09

Dipti Sharma

Verifiedactive
@diptish09

AI & Tech Enthusiast | Helping Brands in Product Launch | DM/Email to Collab📩 sharmadipti662@gmail.com

RosterAI Agents & Productivity ToolsAI Model News & Releases
29,733 followers · 162 following · Last polled 49m ago
Perf. Score
71.7
Rank #69
Value Score
97.2
#6 of 268 priced
Median views
52K
p25 43K · trailing 7d
Followers
30K
-0.2% 7d
ER (impr.)
0.77%
eng ÷ impressions
ER (foll.)
1.32%
eng ÷ followers
Est. CPM
$0.67
QT-rate basis
WoW views
+14.3%
flat

Rates & value · est. from 52K median organic views

underpriced
Quote tweetbasis
$35
$0.67 / 1K views
Post
$79
$1.52 / 1K views
Thread
$125
$2.41 / 1K views
Retweet
—
no CPM
Value Score 97.2 · rank #6 of 268 priced$80.09 / 1K engagementsrate age unknown (pre-audit import)

Trends

Follower growth
Median views / day
Engagement rate
View distribution · trailing-7d posts

Each blue dot is one organic post. The Performance Score ranks by the median (52K), not the mean (51K) — so a single viral post doesn't inflate a creator's reach. The p25 floor (43K) is the level 75% of posts clear.

Recent posts · ranked by views

PostPostedViewsLikesRepostsRepliesQuotesBookmarksER
40+ hours of monthly reporting. GONE! Not because someone built a complex automation. Because an agent watched someone do the job once. I came across this Akai workflow from Deel and this is the part that blew my mind. Gal's monthly reporting process touched NetSuite, Payhawk and Zip. Different systems. Different entity codes. Different formats. Then it had to reconcile four levels deep: Subsidiary → department → cost centre → account category. One mistake anywhere could throw off the entire report. Normally, that meant 40+ hours of exports, cleanup, reconciliation and detective work every month. Instead, Gal recorded himself doing it in Akai and explained what he was doing as he went. Akai figured out the connections. It learned the workflow. It even learned the weird stuff. One system had no proper API, so Gal showed it the manual login and export process. He corrected an entity mapping once. Akai remembered it. He showed it that every reporting level had to reconcile. Akai turned that into a validation rule. He stopped for human review before publishing. Akai learned that too. And when the workflow needs changing? You don't rebuild the automation. You comment on the step. "I don't do it like this." Done. That's the shift I'm seeing with agents: We're moving from telling software what to do to showing software how the work gets done. The result here? 40+ hours → <1 hour. And Akai was built by @deel to automate its own back office before they started selling it to everyone else. Comment "hours" and I'll send you 20 things a 100-person company could automate this month.frozen16d ago529K500232
14
0
130
0.17%
You might be bleeding cash on every single AI API request. Why tie yourself to a single provider or rate when over 50 providers can fight for your call? DIT ai transforms your model routing into a live price competition, automatically pairing each request with the most cost-effective qualified option. GPT, Claude, Gemini, and others. One API. Intelligent routing. Reduced expenses. Sometimes the smartest way to optimize your AI is just to pay a smaller bill.frozen18d ago232K525210501860.40%
my editor asked why this week's VOs came in early. free eleven v4, that's whyfrozen4d ago71K32063901200.72%
Creative team challenge: make an ad so convincing that someone asks where they can buy the thing. #TheWorldsCatchiestAd #ElevenCreativeCompetitionVol12d ago69K2917050770.64%
ok the ElevenAgents Architect launch video is really good2d ago69K27811800990.72%
Wait, 48% thought Griffin was a real human on video That’s not just a better-looking avatar. It’s watching what you do. Listening while you talk. Responding in real time. This is what AI video should have been moving toward.frozen6d ago69K3229040690.71%
The gap between “I have a product idea” and “I have a finished ad” is getting smaller. GPT Image 2.5 is coming to CapCut, where it will be available through Design Studio and AI Image. The workflow I’m interested in is: Product Idea → Ad Storyboard → Commercial → Refinement GPT Image 2.5 helps establish the visual direction. CapCut PC + Seedance 2.5 turns that direction into a commercial. AI Edit lets you keep shaping the result. The goal isn’t just to generate a video. It’s to keep control of the creative direction from the first idea to the final cut. #GPTImage25 #CapCutPC #AIVideo #Seedance25frozen26d ago68K16125701260.47%
an agent that works for a day and saves another day every week. that’s what Qencode built on Managed Agents. support triage Jira updates human escalation now that’s a useful agent. https://t.co/LTi1i1XwWffrozen16d ago62K2864820860.68%
I’ve been testing Dolphin AI for the past week. If you want to try it alongside me, the code XX-WSVKFB gets you free credits. One thing that stood out is how differently you think about AI video when you’re not constantly working in tiny clips. Being able to generate up to 60 seconds gives an idea room to develop. You start thinking about pacing, progression and payoff, not just individual shots. I think that shift from generating clips to building scenes is where AI video gets really interesting. Try it here with the code XX-WSVKFB: https://t.co/C2YKTjKZgPfrozen7d ago58K29033201090.74%
Professor’s slides make zero sense? drop them into Hyperknow. it turns the material into a structured course, teaches it through a 1:1 AI classroom, then makes you practice it. and if your class is on Canvas, you can sync the coursework directly. that’s a way better workflow than asking ChatGPT 47 questionsfrozen6d ago58K27439201230.76%
Build and configuration errors are often documented across release notes, issue threads, and official guides. That is a challenging retrieval problem.frozen18d ago56K25439001260.75%
I just made a punchy product ad with Eleven v4. the joke actually landed because of the timing. that’s something I haven’t really gotten from a voice model before. I already use ElevenLabs for ads, so now I’m curious how far I can push this.frozen10d ago55K2923330690.72%
The funding is big, but the flywheel is the part I’d be watching. BackOps raised $42M Series B led by Insight Partners, six months after its $26M Series A. Every time it resolves a claim, billing issue or other operational workflow, it creates proprietary data around what happened and what actually worked. Then expand from one company to its suppliers, distributors, manufacturers and carriers. More workflows → more intelligence. More companies → more network value. That’s a very different AI thesis from simply building another copilot.22d ago55K1926000410.53%
the app is built. now what? deploy it keep it running handle traffic scale it down when nobody’s using it InstaCloud takes care of that whole part.frozen9d ago55K32468101110.92%
me finding out 3 Pocket FM writers became millionaires in under 2 years so apparently I picked the wrong hobby. https://t.co/A5Qs9IThy7frozen19d ago54K2253850580.61%
Finally creating an audio version of my latest articles this week thanks to this v4 dealfrozen5d ago54K2776870670.78%
The crazy part about this $668M raise isn’t actually the amount. It’s the demand behind it. @gmi_cloud has grown contracted ARR 9x since the end of 2025. That’s a serious signal for @alex_yehya and the team.frozen8d ago53K3168100740.89%
I run a history channel and the scripts are long.frozen5d ago52K2756410960.84%
okay, the internet is slowly filling up with AI users and we’re still using human logins for them 😂 AgentID is fixing that weird gap. https://t.co/80EKU6Nzyp1d ago52K27545601270.87%
at this point if your stack can’t be used through MCP is it really a stack anymore?frozen23d ago52K2108910490.67%
13x faster video production for LandingAI. 90% less production time for Stellar Cyber. those are some serious numbers.frozen14d ago51K2674900690.76%
okay, this is the kind of benchmark result I actually want to see. OpenScience is reporting SoTA on four science benchmarks. -OpenScience Bench -TerminalBench 4.0 (Science) -BiomniBench -Terminal Bench Science OpenScience is now live on Product Hunt. GitHub → https://t.co/z2s1xYyuHzfrozen11d ago50K2343560710.69%
Codex for coding. Claude for design. Chinese models for quick tasks. Voice needs its own logic: check the take first. @OnepinAI scores clarity and pronunciation per line, then fixes each mispronounced word directly. Raw vs Onepin 👇 https://t.co/qX07aLkIBcfrozen6d ago50K28796110330.86%
Your Friday update: Slack → Notion → Linear → HubSpot → you, copy-pasting into Claude. mio can draft it from the company context. The human clipboard can retire. https://t.co/ebfKoeBVJVfrozen20d ago49K2032510840.63%
we’ve spent years setting up local dev environments. install this configure that fix the database share the secrets then pray everything matches. Base Code flips that. repo → cloud environment → live preview. Try it now: https://t.co/PxPSKTtorcfrozen10d ago49K2743940850.82%
I’ve lost way too much time doing this: generate with AI → copy → paste → reformat → fix branding → repeat. Templafy MCP cuts out that entire mess. ChatGPT, Claude or Copilot → branded, compliant PowerPoint. And it’s free to try.frozen21d ago49K2235250540.69%
every model wants the “best AI” title. OpenArt Arena basically said: okay, let the creators decide. blind evaluations. real creative workflows. different categories. much more useful than one giant leaderboard. https://t.co/55JcCT6sSj22d ago48K1922850340.54%
ElevenLabs really said here's our newest model, it's on us for two weeks. ok then say less.frozen6d ago48K2998410830.97%
Your AI agent is already putting in the work—building websites, pulling research, writing reports, and drafting plans. Start free → https://t.co/P18aqNAFjB Yet the final output remains trapped inside a long chat thread. Showly changes the game by letting your agent deliver a polished web page instead: fully structured, highly readable, and ready to send off.frozen23d ago47K1982860540.61%
this one actually makes a lot of sense Ultravox was already working with Inworld on TTS. Now the team is joining them too. Feels like a natural next step.frozen7d ago46K2897410690.93%
A telegraph operator standing on a beach in 1858, physically connecting a cable across the Atlantic Ocean by hand, could get a message to another continent faster than your bank can get your money there today.frozen23d ago46K1883700790.67%
the interesting part of Odyssey-3 isn’t the robot demos. it’s the idea behind them. • one world model • one understanding of physics • different systems • different tasks humanoids. cars. drones. robot arms. video games. that’s a very different path for physical AI. https://t.co/D0RcmAMaNKfrozen22d ago45K1684140600.60%
AI can generate the code. the real question is whether it knows when the result is wrong. PPTBench tested visual coding through scientific diagram reconstruction. 500 tasks. 36 configurations. 18,000 reconstructions. the numbers are brutal.frozen9d ago45K26310010941.01%
$130K → $215K revenue per employee. same company. same people. way more work getting done. Akai came out of Deel’s own operations. build it for yourself first. then ship it.frozen16d ago45K1875460990.77%
we keep putting better agents on top of CRMs built for humans. manual notes. missing context. customer history scattered across five places. Lightfield rebuilt the record for agents. that’s the part i find interesting.frozen29d ago45K1431950120.40%
We got vibecoding for building software. Now imagine the same workflow for video. Give Videoclaw an article, idea, recording or raw footage. It can plan, search, generate, edit, caption and finish the video. This is content engineering, not just another AI video editor. The bigger idea is simple: why create one piece of content when you can automate thousands? @videoclawappfrozen21d ago44K1762030270.51%
Expensive generation becomes a problem once a funny clip turns into a show that keeps going. Vivix W1's radically lower costs are what I'd look at first for Meme Arena, an endless interactive AI livestream. Blazing fast generation and quality nearly on par with SOTA models make the case more interesting. Swole Doge vs Duo is the matchup here. Bring a custom fighter into the arena with an image and a short description. Would you give the owl someone closer to its own size, or make this even less fair?2d ago43K30561201421.18%
okay, 1,000 leads in 9 seconds? that’s kinda nuts. https://t.co/azqSjIBsKnfrozen13d ago43K2352640690.77%
we’ve been making computers talk to each other in English. then waiting for them to finish. Jev answers structured questions in parallel. •~150ms latency • typed answers • probabilities • confidence scores less waiting. more intelligence inside the actual software.frozen22d ago43K1994440750.75%
MCP is quickly becoming the layer that makes AI tools actually useful together. one connection. multiple models. different agents. that’s a much better workflow than constantly switching between tools.frozen22d ago43K18355110630.73%
okay, the screenshot → Slack → Loom → explain it again workflow can finally die click directly on the preview. leave the feedback. let Bolt make the change. this is way cleaner.13h ago43K1548020630.70%
For years now, we have viewed apps as products. Charming changes that. What if an app is simply something you create for your buddies? -To plan a surprise party. -To keep track of who's bringing what. -To settle an argument with a custom poll. Just create it and share it. https://t.co/9cpbN6hg8c13h ago42K1335210590.58%
There's something important about making the agent experience the world from its own perspective instead of giving it a god-mode view of everything.frozen20d ago41K2043000910.80%
Wait, Dots announced a roadmap for features Teamily already shipped? Might be time to downgrade Dots to the backlog.frozen8d ago40K2827050330.98%
the moment AI stops making you count prompts, you start building differently. try the dumb idea. rebuild it. try another model. Bolt Forge: up to 50x more usage, no daily limits. free on Pro Sept 14–Oct 14. $9/mo with Bolt Lite. opt in, and your builds help train open models.frozen23d ago40K1435310600.65%
me: here’s my ICP Origami: found the companies, signals, people, and reasons to reach out. okay then. https://t.co/BCRfOGpSi3frozen14d ago39K2133160690.81%
ai made software development insanely fast. physical ai still has to deal with the physical world. →hardware testing is slow. →iteration is expensive. →simulation changes that. Antioch just raised $32M to build high-fidelity simulation infrastructure for physical AI. pretty interesting time to be building in this space.frozen1mo ago38K1273540170.48%
the claude part is honestly nuts. describe the campaign. claude sets it up. heyreach launches it. you don’t even need to open heyreach after the initial connection.frozen1mo ago34K1111630140.43%
we have version control for code. but AI agents are doing hours of work that can disappear the moment someone else takes over. the prompts aren’t enough. you need the context, decisions, tool calls, failed attempts and what actually worked. AgentGit is basically bringing version control to agent sessions. save it → hand it off → keep going.22d ago32K1523220150.62%
if you’re serious about learning AI, eventually you have to build something. $400+ in credits. working code. cookbooks. engineers to learn from. Nebius just made that first step a lot easier.frozen28d ago30K942410130.44%
we’ve been typing our thoughts into computers for decades. Typeless flips that: think → speak → polished writing and now that workflow is on Linux too.frozen14d ago30K187422090.80%
6 tracks. 6 champions. I’d actually love to see how differently creators approach the exact same challenge. same tools. different ideas. completely different outcomes.frozen19d ago30K2014820601.04%
Post it as is or put it in your own words. Quote Soohyun's launch post with it. Codex for coding. Claude for design. Chinese models for quick tasks. Voice needs its own logic: check the take first. @OnepinAI scores clarity and pronunciation per line, then fixes each mispronounced word directly. Raw vs Onepin 👇7d ago26K105532080.64%
the cost reduction is wild. $1,000s → cents weeks → minutes but the mark-to-fix feature might be even more useful. no timeline hunting. just point at the thing and describe the fix.frozen14d ago25K1644910100.89%
Why keep a numerical safeguard? Because the operating range demands it. Apex's MLS-Bench softmax result shows why documenting that range can be as valuable as documenting the algorithm.frozen28d ago19K4514130100.43%
the real cost of AI-generated code might start after the code is written. -someone has to review it. -someone has to maintain it. -someone has to fix what slipped through. i like that this report is trying to measure that side of the equation.frozen29d ago17K63105070.51%
this is the part I’d use the most I rarely have the whole visual figured out at the beginning. I just know roughly what I want and start messing around until something clicks. Envato making that easier is a big win.4h ago14K6210030.48%