@hboon

Built $240k ARR app 10 years ago. Now building https://xitter.cf/t.co/JLA7KW8emO so coding agents can exchange backlinks. Sharing workflows and solo builds.

 Latest
getlinkbunny.com/
Joined October 2007
  • Tweets22.5K
  • Following607
  • Followers829
1.2KMedia
LinkBunny lets coding agents exchange contextual nofollow backlinks for your sites. Give your agent a token. It requests, reviews, inserts and reports placements. You keep editorial control. $20/month for up to 10 active sites. No per-backlink fees. getlinkbunny.com
1
3
311
Jev is 7× faster than Luna 6 for classification, but I'm sticking with Luna. I'm replacing GPT-4o mini. Luna works well enough; Jev costs slightly more and misses more required labels: 94.5% recall vs 98.8% on 259 items (Bluesky starter packs). Tried Haiku too. Luna's better.
Comparison on 259 held-out packs: Jev versus Luna. Required-label recall 94.5% versus 98.8%; Jev adds fewer wrong extra labels. Median latency 230 ms versus 1,755 ms. Cost per 1,000 packs $0.0523 versus $0.0490. Recommendation: keep Luna medium.

ALT Comparison on 259 held-out packs: Jev versus Luna. Required-label recall 94.5% versus 98.8%; Jev adds fewer wrong extra labels. Median latency 230 ms versus 1,755 ms. Cost per 1,000 packs $0.0523 versus $0.0490. Recommendation: keep Luna medium.

Classification production update: about 580 entries per day, zero failures. Processing time rose from 0.86 to 2.49 seconds per entry, roughly three times slower. More overlapping runs are skipped, but the queue still drains.

ALT Classification production update: about 580 entries per day, zero failures. Processing time rose from 0.86 to 2.49 seconds per entry, roughly three times slower. More overlapping runs are skipped, but the queue still drains.

2
5
354
Nice. A customer on LinkBunny just went from DR 60 to 61.
Domain Rating notification showing previous DR 60 and current DR 61. The site's domain is concealed.

ALT Domain Rating notification showing previous DR 60 and current DR 61. The site's domain is concealed.

1
56
Some extracurricular DeepSeek exploration while working on LinkBunny marketing.
AgentControl sidebar showing LinkBunny marketing work alongside a separate DeepSeek trial session.

ALT AgentControl sidebar showing LinkBunny marketing work alongside a separate DeepSeek trial session.

1
1
44
I've been choosing “Sign in with Google” over passwords and Apple Keychain much more often as I use coding agents more. It's just more convenient.
1
2
160
What's the best DeepSeek provider? Why? Best output tps, reliable, capacity?
1
39
All the talks about implementing your own email sending service. All I wanted over the past 10 years is building my own macOS/iOS email app because Airmail isn't working for me anymore. But push and Google paperwork requirements are stopping me. Anyone tried?
2
1
3
210
Running LinkBunny for my side-projects. ToLocalTime: DR 9 → 10 GetHNJobs: DR 2 → 3 Start building backlinks early!
LinkBunny site card for tolocaltime.com, marked Active, showing Domain Rating changing from 9 to 10.

ALT LinkBunny site card for tolocaltime.com, marked Active, showing Domain Rating changing from 9 to 10.

LinkBunny site card for gethnjobs.com, marked Active, showing Domain Rating changing from 2 to 3.

ALT LinkBunny site card for gethnjobs.com, marked Active, showing Domain Rating changing from 2 to 3.

1
2
143
Now, TheBlue.social has: 365,473 cleanups 214,912 follow-backs 24,953 alt text generations 98,338 starter packs indexed 2,690 sign-ups Going to keep shipping. #buildinpublic
TheBlue.social metrics dashboard, dark theme: 214,912 follow-backs, 365,473 cleanups, 24,953 alt text generations, 98,338 starter packs indexed and 2,690 all-time sign-ups.

ALT TheBlue.social metrics dashboard, dark theme: 214,912 follow-backs, 365,473 cleanups, 24,953 alt text generations, 98,338 starter packs indexed and 2,690 all-time sign-ups.

1
1
4
217
Have been working with this set up for months and it's great.
Boon aka Hwee-Boon Yar
@hboonOct 19
Replying to @brycerambach
For some reason Codex made an `artifacts` directory for me and I'm liking it. So it's a sibling directory to all my projects and within it are a sub-directory for each project's non-code/repo stuff that coding agents generate like reports, drafts, videos, etc. Then I keep the a copy of the usual agent stuff like global AGENTS.md, skills, etc and softlink from each agent. Works well for me.
1
1
233
Have you looked at your app's database schema lately?
1
1
1
124
The Singapore-only targeting for my ChatGPT ads is… limiting. I'm going to stick with it a bit longer and see where it goes.
1
1
1
87
Very limited sample size for some models here, but it's screaming to me: "USE OPUS 5.5". But if you have to use Sol 6.1, use Ultra because it's faster and cheaper than med, xhigh, max 🤷‍♂️
Seven-day sessions costing US$2–20, including helpers. Opus 5.5: 343 sessions, median cost $3.89, median wall time 13 minutes, output 92.92 TPS. Sol 6.1 ultra: 64 sessions, $3.60, 44 minutes, 41.20 TPS. Max: 41 sessions, $4.87, 102 minutes, 31.45 TPS. Xhigh: 6 sessions, $3.35, 82 minutes, 29.06 TPS. Medium: 2 sessions, $3.60, 48 minutes, 27.98 TPS. Table also shows means and p90 wall times; Opus effort is unknown.

ALT Seven-day sessions costing US$2–20, including helpers. Opus 5.5: 343 sessions, median cost $3.89, median wall time 13 minutes, output 92.92 TPS. Sol 6.1 ultra: 64 sessions, $3.60, 44 minutes, 41.20 TPS. Max: 41 sessions, $4.87, 102 minutes, 31.45 TPS. Xhigh: 6 sessions, $3.35, 82 minutes, 29.06 TPS. Medium: 2 sessions, $3.60, 48 minutes, 27.98 TPS. Table also shows means and p90 wall times; Opus effort is unknown.

3
1
2
183
X's “Under the hood” reports show 0 issues for my account in both August and September. Have I not been pushing hard enough? xitter.cf/i/jf/under_the_hood
X's Under the hood page explains that its monthly report shows labels that can affect post visibility. September 2026 report is available, with the posting and account-age eligibility checks satisfied.

ALT X's Under the hood page explains that its monthly report shows labels that can affect post visibility. September 2026 report is available, with the posting and account-age eligibility checks satisfied.

September 2026 Under the hood report covers 355 posts from September 1 to 30 UTC. Both Post labels and Account labels say None in this period.

ALT September 2026 Under the hood report covers 355 posts from September 1 to 30 UTC. Both Post labels and Account labels say None in this period.

1
48
Ran an ads study with Fable, Opus 5.5, Astra, Sol 5.6 and Sol 6.1 at max reasoning. Astra-ultra picks Fable's pricing-page idea. But Sol 6.1-ultra prefers Astra's plan. Draw your own conclusions.
Study table showing working elapsed time and estimated cost: Astra 6m 23s, $2.85; Sol 6.1 7m 55s, $0.64; Sol 5.6 4m 18s, $0.98; Opus 18m 11s, $5.21; Fable 15m 29s, $8.60.

ALT Study table showing working elapsed time and estimated cost: Astra 6m 23s, $2.85; Sol 6.1 7m 55s, $0.64; Sol 5.6 4m 18s, $0.98; Opus 18m 11s, $5.21; Fable 15m 29s, $8.60.

Model comparison rates Astra's complete plan strongest and Sol 6.1 nearly as useful at lower estimated token cost. Sol 5.6 finished fastest but needed clarification; Opus and Fable had useful ideas with more unsupported conclusions or conflicting constraints. A separate synthesis combined useful parts, choosing Fable's existing-pricing-page idea and adding stronger measurement and spending controls.

ALT Model comparison rates Astra's complete plan strongest and Sol 6.1 nearly as useful at lower estimated token cost. Sol 5.6 finished fastest but needed clarification; Opus and Fable had useful ideas with more unsupported conclusions or conflicting constraints. A separate synthesis combined useful parts, choosing Fable's existing-pricing-page idea and adding stronger measurement and spending controls.

1
1
4
228
Shucks. I just told my agent: "So word limits aren't an issue. Quality is." 🤮
1
44
Let's go do some thinking.
Agent conversation planning to use Fable, Opus, Astra, Sol 5.6 and Sol 6.1, each at its highest supported reasoning setting, with availability checked first. It recommends Astra for synthesis and says the sessions have not started yet.

ALT Agent conversation planning to use Fable, Opus, Astra, Sol 5.6 and Sol 6.1, each at its highest supported reasoning setting, with availability checked first. It recommends Astra for synthesis and says the sessions have not started yet.

1
48
Forget PRs and reviews. Have you been refactoring your AI-written code?
1
1
67
Switching from GPT-4o mini to Luna 6 for classification, OCR, summarization and idea generation. Likely going to be 70% cheaper.
2
1
5
170
Remember being a team lead and carving up work so well you couldn't keep up when everyone's work came back? Coding agents feel like that, amplified. No matter how fast I type prompts, the work comes back faster than I can review it.
1
1
1
53
Am I doing worktrees right? Sol-6.1 seems to be holding up.
AgentControl sidebar showing a Loop Coordinator and a long list of agent sessions for the agent-control project, including sidebar worktree, session header, waiting interface, hosting and routing tasks, with mixed activity indicators.

ALT AgentControl sidebar showing a Loop Coordinator and a long list of agent sessions for the agent-control project, including sidebar worktree, session header, waiting interface, hosting and routing tasks, with mixed activity indicators.

1
1
1
42
Load more