Who maxes out 5 AI subscriptions?
Executive Summary
A potluck episode of listener questions. The through-line is the working conditions of AI-era development: what self-promotion looks like after blogs (opinions, and authenticity as the new gate — AI-fakeable output is discounted while traceable human work stands out), what to do while agents run (document UI problems in detail rather than idle; tighter loops beat long turns), what employee AI budgets actually look like ($200–$500 a month for most, enterprise token billing a worse deal, restraint a skill), and how transactional email should be built (standard APIs or SMTP over proprietary ones, because deliverability is fragile). On agent control, Scott lays out deterministic guardrails — harness hooks that block an rm, deny lists, linters, frame-rate tests, routing automations like his Jellyfin status check, agents.md rules, container isolation — plus Wes's escape-key habit, and the steering comparison: stop by default, let Claude Code pick up mid-run, Codex's queue-next-turn is the current best, OpenCode has none. Adoption advice for an engineering lead: skip the one-prompt stunts; use walkthroughs, a Sentry-style internal Slack channel, and screen recordings. On agent UIs, chat-shaped super-apps (Codex-in-ChatGPT, Grokbot, Muse) are consolidating while custom builds break weekly. The most personal exchange confronts lost motivation: AI coding feels like Game Genie cheats, and Scott's recovery is moving joy up a layer — strategy, refinement, and improving one's own process and harness, so code becomes the result of process improvement. The closing architecture question gets a middle-path verdict — custom tooling around a harness yes, the harness itself no — under a shelf-life warning: much of what they learned in eight months is already obsolete, so audit which installed pieces still earn their token churn.
Chapters & Key Takeaways
10 takeawaysOpinions are the post-blog differentiator, and authenticity is the gate: AI-fakeable output (slop repos, formulaic video) gets discounted, while traceable human work — like a Netflix engineer live-sharing a rebuild — is what a hiring manager rewards.
Yapping is the new blogging, but there is always room for people who know what they are talking about and can communicate it well.
Agent wait time is specification time: a cheap issue-filing agent plus an orchestrator turns waiting into a detailed UI-problem backlog, replacing vague 'make the UI better' prompts.
Most employers now cover $200–$500 a month of employee AI use; enterprise token billing is the worse deal, and the $200 subscriptions (Wes measured $1,100 of weekly tokens on them) are the best value.
Send transactional email through standard APIs or SMTP — a 2% spam-trap rate is real lost revenue, and standards keep provider switching possible.
Keeping agents on the rails is layered: deterministic hooks that block an rm, deny lists, linters and frame-rate tests, routing automations that hand the agent state and instructions, agents.md rules, and container isolation — with the honest caveat that derailment frequent enough means the human is yoloing prompts.
Steering mid-run: stop by default because interventions are urgent; Claude Code is the exception worth letting pick up mid-run; Codex's queue-next-turn is the best current implementation; OpenCode has none.
For team adoption, one-prompt stunt demos backfire; walkthroughs of real workflows, a shared internal channel where senior engineers trade criticism and tips, and screen recordings of actual debugging work.
AI coding feels like Game Genie cheats — the challenge is gone in five minutes — and recovering motivation means moving the joy up a layer to strategy, refinement, and improving one's own process, so code becomes the result of process improvement.
Build custom tooling around a harness, never the harness itself; and treat any answer as perishable — much of eight months of AI-tooling lessons is already obsolete (Context7 may be unnecessary now that search is good).
Deep Read
What Does Sustainable Work With Coding Agents Look Like?
On Syntax, the hosts put numbers on healthy AI usage — most employers cover $200–$500 a month, $200 subscriptions beat enterprise token billing, and Wes metered $1,100 of weekly tokens on his — then add the guardrails, the wait-time discipline, and the honest answer to lost motivation.
Read the full articleInterview Highlights
10 exchanges- Q
What is the modern-day equivalent of blogging for developers and designers who want to stand out?
Scott Tolinski asked01:36AScott Tolinski, Wes Bos answeredBoth hosts answer: opinions are the new differentiator. Scott says audiences now reward people who took the time to figure something out and formed a view, not the high-level demo; distilling real experience cohesively is what promotes a person. Blogs still work — Hacker News is still mostly text — and short-form video is worth practicing, but Wes's summary is that 'yapping is the new blogging'. Wes adds that the barrier is now authenticity: AI-fakeable output (slop GitHub repos, formulaic YouTube takes) is discounted, while human, traceable work — like a Netflix engineer he follows sharing his Hugging Face rebuild bit by bit on Twitter with text and photos — is exactly what would put someone at the top of his hiring list. Their closing advice: do interesting things, put them out there, and let the opinions stand out to employers.
- Q
What should you do while your coding agent runs for minutes at a time?
Scott Tolinski asked08:33AScott Tolinski, Wes Bos answeredScott's honest answer: he puts another agent on a different task, then another, until 'your brain fries' — while noting push-ups or a walk would probably be better. Wes prefers tighter loops: an agent running ten minutes already annoys him, and even two-to-three-minute waits break flow, which is why people say the flow state is gone. Scott's actual method for the wait: inspect the application and write down, in detail, everything wrong with the UI. He runs an orchestrator agent plus a cheap 'GitHub issue' agent, brain-dumps problems into issues while the orchestrator works, and finds that documenting what needs fixing — researching design patterns and distilling them — takes longer than a single agent turn and builds a backlog so the next pickup is never 'make the UI better, more modern'. Wes frames the underlying shift: AI coding is refining a statue that starts close to done, so that process needs the eye and intent to refine.
- Q
How much should an employer cover for employee AI usage, and how do budgets actually look?
Scott Tolinski asked14:01AWes Bos answeredWes reports that company stances changed in the last six months: after the 'magic box, prompt forever' phase produced runaway spend, slop, review load, and security issues, most employers he polled on Twitter now land between $200 and $500 a month — a handful unlimited, some $200 for one sub, some $500 covering several subs that people burn through in sequence — while enterprise token billing at Anthropic and Claude-scale vendors is a much worse deal than the $200 subscriptions. He treats budgeting as a skill developers must now learn, like performance tuning once was: you can easily spend a thousand dollars a day building features, so restraint matters, and a limited budget with the ability to scale up — 'boss, we're rewriting in Rust, I need 20 grand of tokens' — is the sane pattern. He tracks his own usage with a Mac app, Codex Bar, wired to all his subscriptions; it showed $1,100 of tokens in a week he did not pay $1,100 for, which is why he calls the $200 subs by far the best deal — get one if your company pays.
- Q
How should you build the transactional emails your app sends in 2026?
Scott Tolinski asked20:31AScott Tolinski, Wes Bos answeredScott always rolls his own templates — trivial in the agent era — unless the platform (he names PocketBase) builds them in; his pain point is cost, since he hates paying for occasional transactional email, and he currently sends through Resend's API while eyeing Postmark (free at 100 emails a month) and Amazon SES's pay-per-email for low-volume side projects. Wes's stack is React's render-to-string with inlined CSS for reusable templates, sent over plain SMTP through any provider like SendGrid, precisely so no single company owns his pipeline. Both converge on the portability rule: always send through standard APIs or SMTP, because deliverability is fragile — if 2% of mail suddenly lands in spam, that is real lost revenue, which is why Resend is such a big company — and a standards-based sender can swap providers the moment it happens.
- Q
How do you keep a coding agent from going off the rails — and would you say it can be done reliably?
Scott Tolinski asked25:26AScott Tolinski, Wes Bos answeredScott's mechanism list: harness hooks like Claude Code's can deterministically intercept a dangerous call — detect an rm attempt, block it, and tell the agent what to do instead — plus deny lists, linters, and tests; he wrote frame-rate tests that spin up an agent browser after each implemented feature and fail under about 55 fps. For routing he uses deterministic automations: when he asked an agent why his Jellyfin was down and it spun up an unrequested replacement container, he wired a status-check automation so the agent now gets state plus instructions instead of improvising. Rules in agents.md ('don't write fallbacks for absolutely everything') curb feature bloat, isolation and containers bound the blast radius (their producer CJ's recommendation), and Wes adds the human habit: he watches the output and spams escape the moment the agent wanders. Scott's caveat answers the interview question: there are still times the agent does what it wants, so if it derails constantly, that is on you — you are yoloing the prompts too much and need to stay closer.
- Q
What is the best way to steer a coding agent mid-run — stop it, queue a message, or use by-the-way features?
Wes Bos asked30:25AScott Tolinski answeredScott has not found a steering pattern he loves. He dislikes by-the-way features — that queued turn often takes too long — and his rule is that stepping in is usually urgent, so he defaults to stopping the run regardless. The one exception he names is Claude Code, which he lets pick up mid-run because it handles the injection particularly well; even then he does not mind hitting stop, since the agent usually is not doing damage and the next turn, twenty or thirty seconds away, is a fine place to add context or change course. Wes closes unsatisfied but decisive: none of them are great, and Codex's queue-next-turn behavior is his favorite so far.
- Q
How do you get engineers interested in agentic development when interest on the team is low?
Scott Tolinski asked32:48AScott Tolinski, Wes Bos answeredScott warns that some engineers carry an ideological anger toward AI tooling and will not be moved, and that many resisters have not registered how much better the models have become; the 'look at this crazy one-prompt thing' demo actively hurts. What works instead: lunch-and-learn sessions showing real workflows — how it improves your development flow and your code — rather than stunts; relating to the person, since this is a people problem about the wall they have built. Wes adds that leadership must visibly share how they themselves use the tool, ideally in a company Slack channel like Sentry's, where top engineers trade questions, criticism, and tips; he loves that channel. Scott's last suggestion is internal screen recordings of an actual debugging process — the bug, the steps, the benefit at the other end — even if that format does poorly publicly.
- Q
What interface will non-coding AI agents get — Telegram-style chat, chat super-apps, or purpose-built UIs?
Scott Tolinski asked35:52AScott Tolinski, Wes Bos answeredScott says everyone is hunting the golden-goose UI right now. Codex has merged into ChatGPT as one app, and the two big general-purpose agents he names — Grokbot and Facebook's Muse — are chat applications with few knobs, aimed at daily tasks rather than coding. His read: plain Telegram back-and-forth is probably not it, because you need somewhat more control over the UI — but not so much that it overwhelms, which is why Grokbot and Muse lean chat-heavy. Wes testifies from the custom end: his own purpose-built setup is great when it works but 'breaks all the gosh darn time' and earns a weekly 'the AI is not working again' message from his wife; third-party options like the Hermes desktop app are really good, though. Scott's position lands in between: he still runs his custom one, and it is kind of annoying.
- Q
How do you stay motivated to code when AI has drained the excitement from it?
Scott Tolinski asked38:43AScott Tolinski answeredScott answers from his own struggle: AI coding is like playing a game with Game Genie — the cheats are fun for five minutes, and then the challenge is gone, and coding's deep digging is exactly the part you lose when the workflow becomes 'agent, fix this'. He also acknowledges AI coding is more exhausting for other reasons — broken flow and what their listener mental-health survey surfaced. What restored his drive was moving up a layer: finding the joy in strategy, deep UI refinement, and larger structures and architectures — and in improving his own process, harness, and skills, so that the code becomes a result of an improved process rather than something improved by his hands directly. Wes's whole reply: 'I agree. Well put.'
- Q
Should a company build its own all-in-one AI harness, or keep humans as the orchestrator with small per-problem agents?
Scott Tolinski asked44:16AScott Tolinski, Wes Bos answeredScott grants that harnesses are genuinely hard — caching, model routing, efficiency — which is why harness experts exist. But he has never considered rolling his own despite his build-it-yourself instincts: custom tooling around a harness makes sense, building the harness itself does not. His verdict on B Murf's either/or: the answer is in the middle — the company went left, the listener went right, and sadly the truth is between the monolithic harness and human-orchestrated micro-agents. Wes's addendum undercuts the shelf life of any answer: much of what they learned in the last eight months already does not matter — he now doubts they still need the Context7 MCP for docs because search got so good — and Scott seconds it with his 3D-printable rendering story (a library that collapsed a year ago now works), adding that installed skills often just add token churn, so audit which pieces actually help.