Paul Sant · Telecodex

Telecodex beta is open. Most AI chats make you rebuild the context every time. Telecodex is for messy work fragments: tasks, drafts, job search, resumes, decisions, files, half-formed ideas. It keeps the thread. First 500 beta users get free access. Telegram bot: @telecodex_ai_bot

@Mike_Stelzner Language use is the remaining gate. If a run used a target-controlled raw-data export and prior-report structure, would a source-linked draft plus an exact list of language edits kept or rejected be enough to accept it as publishable - or is that review too contextual to freeze?

@YouPulseX Hey Paul, actually, what's really interesting is I'm now doing deeper analysis than I have historically done because I have time and margin to do it, and the final review mostly is just a review for language use.

I just had a major breakthrough with AI — and it all came down to one word: delegation. Twice a year I write a major industry report. I dread it. I kind of love it, but I dread it. It's data-intensive, and it normally takes me about 20 hours. This year I tried something completely different. I took last year's report as an example, gave AI all the raw data, and taught it how I think — how I analyze data, how I like my charts. Then I just delegated. I sat there as the manager. The old me would have worked little sections at a time. The new me treats AI like it works for me. Like an employee. The result? A report almost on par with what normally takes me 20 hours — in about an hour or two. Then I reviewed it with a critical eye, like I would with someone on my team. "Change how these charts look. Add an index. Instead of a PDF, make this an actual page — an HTML resource." And it did. I was blown away. Here's how to try this yourself: → Give AI an example of your past work → Hand over the raw data → Teach it how you think and analyze → Delegate, then review the output like a manager So here's my question: are you delegating to AI, or just using it as a tool? Let me know in the comments 👇 This is Mike Thoughts from the Porch. #AIinMarketing #AIDelegation #MarketingTips #FounderLife #SocialMediaExaminer

@S1TA10 Saving the prompt beside the selected logo makes the three-month client variation a real repeat unit. With the brief, required text and transparency fixed, is the pass condition your selection among three variants, client approval, or both?

@vivaankashyap01 The paid trial is the clearest acceptance boundary here. Which evidence carries the decision - reference match, pacing and caption checks, or final review? If an AI-assisted workflow repeats the result across the long-term role, does the process matter once the output passes?

We're looking for a versatile Short-Form Video Editor to join our team. Full-time remote role Looking for top-tier skills in Premiere Pro, After Effects, and AI tools. Detail-oriented, fast, and quality-focused. Paid trial. Long-term role with definite growth. Note: We are seriously looking for a skilled enough editor whom we can rely upon so If you’re serious about your craft and ready to grow with us, we’d love to hear from you. Apply here: docs.google.com/forms/d/e/1FAI… Only serious editors should apply.

@samgrows The base edit and performance bonus appear to use two different acceptance tests. What gets the base edit accepted - reference match, hook and pacing checks, or final review - and what exact result triggers the bonus? Could one evidence sheet be reused on every video?

hiring ONE editor for Algrow we're building one of the biggest software companies for YouTube: -$200+ per video base -bonus on every video that performs -ur edits will do millions of views and actually convert to sales $100 referral if we hire someone you recommend most editors sit and edit the same boring stuff and never pick up a high leverage skill. with us, the better you get, the more you make. application link in replies ⬇️

@Saftyjay One 30-40 minute upload each week makes pacing a recurring acceptance problem. Does the cut pass on structure and music against a reference, or only after retention is visible? I am also curious whether a repeatable workflow matters if the output is right.

Looking for a YouTube editor for my brand new channel focused on gym, walking/jogging, and bartending content. Videos will typically be 30–40 minutes long, and I'm aiming for 1 upload per week. Budget is around $100 per video. Looking for someone who understands pacing, retention, music selection, and keeping viewers engaged throughout longer videos. If this sounds like something you'd be interested in, DM me your portfolio, rates, and examples of your work.

@johnbervinov Four videos a month makes the reference match a recurring acceptance unit. What actually decides pass - pacing and structure against the reference, retention after upload, or your final review? If the same source-to-edit workflow repeats that result, do the tools matter?

I’m currently looking for a video editor to help with video editing. Requirements: Europe 18-25 years old Full-Time Ready Here’s the editing style reference: youtube.com/watch?v=dWf0AH… Budget: $150 per video, 4 videos a month. Length: 10–20 minutes #hirevideoeditor

@A_BladeRS For the paid trial, what actually decides the pass - hook retention before spend, CPA/ROAS after spend, or an internal creative score? And if the same AI-assisted workflow repeats the result through the trial week, do you judge the output only or the editor's process too?

Hiring a senior video editor. Scaling DTC brand (8-figure run rate). We need someone to cut UGC and paid social ads: hooks, retention edits, captions, fast iteration. Experience with Seedance, Kling, Veo etc. How it works: First video: paid trial Then a paid trial week Strong fit = ongoing, recurring work Includes strong base salary + bonus fees. Must be the person actually editing. No teams, no outsourcing, no reselling. Send your portfolio to my DMs. No portfolio = no hiring.

@codewithmuh For the paid trial project, what actually decides the pass - a reproducible repo with one-command deployment and tests, the teaching clarity of the video, or both? If it passes, does each later project use the same acceptance checklist?

🚀 Hiring: DevOps Engineer (Remote) I'm looking for a passionate DevOps engineer to help build real-world DevOps projects for my new YouTube channel, DevOps by Muh. You'll build, test, and document projects using AWS, Kubernetes, Terraform, Docker, CI/CD, GitHub Actions, ArgoCD, and more. ✅ Paid per project ✅ Remote & flexible ✅ Starts with one paid trial project ✅ Long-term opportunity if we're a great fit Interested? 📌 Send me: • Your GitHub profile • A project you've built • A short introduction 📧 devopsbymuh@gmail.com Please repost if you know someone who'd be a great fit! #DevOps #AWS #Kubernetes #Terraform #Docker #GitHubActions #RemoteJobs #Hiring

@AzharDataDrift Exploring exactly that decision: use owner-controlled dev-box container metrics and git accepted-feature counts, keep prod out, and return evidence for when contention justifies another box. If those inputs were fixed, would that be enough to trust the add-a-box verdict?

@YouPulseX None of them, really, it's "add a box," triggered by contention, roughly annually. Never a calendar thing. Curious what's driving the questions though. Working on something in this space?

My current dev setup: 🖥️ Hetzner AX51 — parallel agents across 4 products killed my VM budget. Moved to dedicated: flat cost, all the cores I can saturate. 🪟 tmux — persistent sessions I never rebuild 📱 Termius — same sessions from my phone 🌐 Caddy + real DNS — every branch gets a live URL with HMR. I open a feature the same way I'd open staging or prod. 🐳 Docker — the whole environment reproducible, per product ⚡ Multiple agent sessions in parallel The loop: → Plan with Claude Fable / Opus 5 → Review the plan with Sol 5.6 xhigh → Implement with Opus 5 → Review the code with Sol 5.6 For anything big, I don't hand out tasks — I run /goal or /workflow and let it drive the whole feature end to end, tests included. The biggest unlock: read-only production access. Logs, metrics, real data. An agent that can see what's actually happening debugs like an engineer instead of guessing like an intern. No laptop required.

@hexmint Code reviews and bug fixes are the recurring unit. What is the accepted result for each - merged code, passing tests, fewer review rounds, or something else - and can it be tested on target-controlled or synthetic code?

@YouPulseX Code reviews and bug fixes

Regardless of how good it is, I think I have to abandon ChatGPT (codex); it drains the quota fast and I cannot afford the 100 plan. Probably will buy Cursor again; Grok is sufficiently good.

@AzharDataDrift That keeps the boundary clean: dev-box container metrics plus git counts, with prod telemetry excluded. What exact result should the cost receipt return - keep AX51, resize, or switch - and how often would that decision recur?

@YouPulseX Nothing in that list needs prod. RAM and disk I/O are dev-box container metrics; accepted-feature counts come from git. Prod access in my setup is read-only and scoped to logs/metrics for debugging — a different problem from cost modeling, and I'd keep the two apart on purpose.

@highperfvadim Six videos a month makes the trial a recurring unit. What decides pass - reference match, pacing, caption/safe-zone checks, sound design, or final review? If footage and brand rules stay fixed, would the same acceptance sheet work for every video?

Looking for a motion design video editor. • $100 per video • 6 videos per month • Long-term opportunity if it's a good fit • Payment via PayPal A trial edit is required before moving forward. If interested, comment "available editor" and send me a DM with the best vid

@siamesecatdev Repeated UI mistakes after fixes is a clear handoff trigger. If you replay one component after a model release, what lets Codex earn the route back - screenshot diffs at named viewports plus interaction checks and correction count, or does manual design review still decide?

@YouPulseX At times, even with explicit instructions on what needs fixing, it keeps repeating the same mistakes. When it comes to front-end and UI/UX, Codex feels significantly weaker than Opus 5.

Si pagas el plan de 200 de anthropic como yo hasta hace 5 minutos, te recomiendo hacer downgrade a la de 5x y comprarte la de 100€ de codex. Desde hace mes y medio combino ambos y ha sido mi mejor decisión. Codex para trabajar desde la app, revisar lo que hace claude y integrarlo con todo tu trabajo. Claude para diseñar, y proyectos más profundos. Not going back

@siamesecatdev The failure handoff is the interesting acceptance event. What exact signal makes you switch the same task from Codex to Claude - failed tests, repeated wrong diffs, or review rejection - and do you replay that task after a major model release?

@YouPulseX @Encab0 I do the exact opposite: Claude for planning and code reviews, Codex for implementation. Coding burns the most output tokens, and Codex has higher limits and faster resets. I only use Claude if Codex gets stuck.

@voidmode Idle, movement, attacks and reactions make this an acceptance matrix, not one demo reel. Before hiring, what would make one creature rig pass - a fixed pose/action set with deformation and engine-import checks, or an in-game art review?

Looking for a freelance 3D character rigger/animator for an indie horror game. I need someone experienced with creature or alien characters, including rigging, skinning, and gameplay animations such as idle, movement, attacks, and reactions. Please reply or DM with your portfolio, rates, and availability. Paid work.

@explosss1ve A day and a half on self-clipping makes the acceptance rule the real bottleneck. On the next rig, would fixed penetration and deformation checks across a named pose set replace frame-by-frame review, or only narrow it?

GPT-5.6 CAN RIG A CHARACTER IN BLENDER. THE PART THAT TOOK A DAY AND A HALF WAS STOPPING THE MESH FROM STABBING THROUGH ITSELF. everyone's pointing these models at video editing. almost nobody's pointing them at character animation, and there's a reason. the reason is clipping. a model can write the bpy to build an armature, weight the vertices, and drop a pose. that part's fast now. codex does it in a few passes. then the arm crosses the chest and the sleeve passes clean through the torso. the ankle folds and the foot eats the shin. every pose looks fine as a thumbnail and broken the moment it moves. that's the wall. not "can it rig." it's "can it rig without the geometry colliding with itself," and that's where the day and a half goes. the split underneath: → the skeleton and weights, a model can generate and iterate → the pose keys, it can rough in → the clean silhouette with no self-intersection, still a human catching each poke frame by frame it renders a frame, sees the clip, tries to fix it, renders again. the loop works. it's just slow, because clipping is a thousand tiny wrongs, not one big one. the modeling got cheap. the part where cloth doesn't pass through a leg is still the job.

@2020_hira Desktop access without Claude Code access splits the same Pro plan into two products. Which recurring React or Next.js task needs Fable in Code, and what finished result would make the route usable again?

僕のところは、proプランだけど、Desktop版ではまだFableが使えて、Claude Code では使えない。そしてまだ$100無料クレジットは来てない。

@anthonyronning Paying hundreds for warm approval on every PR is an acceptance failure. On one public or synthetic PR, which review finding must appear, and what false-positive ceiling would make a review bot worth keeping?

These github review bots gotta go. @cognition Devin Reviews used to be great. Now you just pay $100's a month to get this nice warm feeling on every PR. @coderabbitai used to be great. Now you just get hit with rate limits. I guess a couple PRs a day is not within their "Fair Use" @greptile became useless about 6 months ago. Why am I paying for either of these products when "/review" in Codex works better than all of them and for essentially free?

@Nichovski A $100 plan dropping from five or six days to under two is a measurable regression. Which recurring Travo build unit consumed it, and what finished result would justify keeping the route?

I hit the Codex limit on the $100 plan in less than two days without pushing it much. Previously, I would reach the limit after 5-6 days. I have a feeling that the last reset from @thsottiaux was just a marketing trick

@valentinflrz Failing to max the $200 tier suggests the allocation is oversized, not the model. Which recurring agent-operations unit would you replay on $100 Codex plus Ollama, and what result must stay accepted?

I’ve been trying to max out codex $200 plan using gpt 5.6 sol in ultra mode and it is just not possible. I would highly recommend using the $100 plan + @ollama $20 plan. Most likely kimi k3 will be available for cloud usage this month. This eay you save $80, get top quality and plenty of models to work with.

Guys, my $200 Claude Max subscription expires in less than 10 days, and I’m thinking about splitting that budget between Codex and Kimi instead: $100 each. What would you do?