One blog post, three platform variants, 1.1 cents. My AI is my content team now.

If you're a developer or founder shipping a product, you've probably had this conversation with yourself: "I should be on LinkedIn. I should be on X. I should be on Instagram. But I don't want to be. And the way each of those platforms actually rewards content is different enough that I can't just paste the same paragraph everywhere and call it a day."

Turns out that gap — between the blog post you actually wrote and the three platform-native versions that would make it land — is exactly the kind of thing an AI agent handles well.

I gave one of my published Reality Router blog posts (the receipt-vision story from a few days ago) to an agent and told it: write me a LinkedIn version, an X thread, and an Instagram caption. Each has to respect the voice and constraints of its platform. Also: for Instagram, propose the visuals I'd shoot.

Three minutes eleven seconds later, I had all three. Cost: $0.011.

What the agent produced

The LinkedIn version opens with a professional-but-human hook, front-loads the "why this matters" value, and lands the technical detail as a payoff rather than a lead. Nine short paragraphs (LinkedIn cuts off after three lines with a "see more" — so the first two paragraphs have to earn the click). Three hashtags at the bottom, none in the body. First line was:

"My AI agent was handed a task it couldn't possibly do. It did it anyway — for three cents."

The X thread was ten tweets, each under 280 characters (the agent counted). Tweet 1 opened cold — no context assumed, since Twitter readers arrive with none. Every subsequent tweet built on the previous one and broke at natural idea boundaries, not mid-sentence. Sample from tweet 5:

"It didn't stop. It found rapidocr-onnxruntime — a pure-Python OCR library that installs without root. No permissions drama, just a pip install."

The final tweet was the CTA + hashtags — the only place where hashtags belong on X.

The Instagram caption was 800 characters — inside the sweet spot of "long enough to give real value, short enough to actually get read" — with four emoji sprinkled naturally through the copy (not clumped at the top like a spam post). Twelve hashtags on the final lines. Then a section I especially cared about: ## visual suggestions, with three specific shoot ideas:

"Photo: a café receipt lying next to a laptop showing a JSON code snippet — the 'human paper → structured data' story in one frame."

"Carousel: swipe through the pipeline — the PNG receipt, the failed first attempt in the terminal, the pip install moment, and the final JSON output."

"Reel: fast screen-recording of the agent probing the environment, installing the OCR package, and producing the JSON, with the three-cent cost line as the payoff."

That last one — the reel concept — is the kind of specific creative direction you'd normally pay a content strategist $200 for.

The receipt

Just over one cent for three platform-native rewrites plus a visual concept brief. The same seven agent calls sent to any of the flagship models would have cost a quarter to a dollar — same three deliverables. The economics of being on three platforms went from "hire someone" or "burn a whole afternoon" to "run it while you make coffee, at flagship-model quality for a rounding-error price."

Why the platform-specific constraints matter

The reason people who cross-post naively get low engagement isn't that their content is bad — it's that each platform rewards a different shape of content:

• LinkedIn rewards professional-voice value delivered in a scannable structure. Long paragraphs kill you. So does hashtag spam in the body. The "see more" cutoff is a hard test at line three.
• X threads reward specificity, momentum, and stand-alone tweets. Each tweet has to work if it's the only one someone reads. Hashtags mid-thread look amateur.
• Instagram rewards emotion, aesthetic setup, and a caption that reads like a caption rather than a press release. Emoji land, but only if they're rhythmic. And crucially: the visual is 80% of the post, so an Instagram plan without a visual concept is a broken plan.

The agent respected all of that. Not because I told it to — I did, briefly — but because the constraints are baked into the platform-appropriate copy patterns any competent LLM has seen a hundred thousand times.

What the router did

Seven agent calls, split five to DeepSeek's fast cloud model, two to gpt-4o-mini for the harder reasoning turns (checking character counts, verifying hashtag placement rules, deciding what to cut vs keep for each platform). No premium tier needed for a task this shape.

That's what "calibrated routing" looks like on a real workflow: right-sized model per call, no thinking required from you about which tier to use, the difference in cost showing up as basically free.

Try it yourself

Reality Router is open source. Point OpenCode (or Aider, or any OpenAI-compatible tool) at it, hand it your last blog post, and see what your content team looks like when it's an AI agent that does everything except the hard part — which is deciding what to write about in the first place.

→ realityrouter.dev

Source post used for this test: the receipt-extraction story from the Reality Router blog. The agent's LinkedIn/X/Instagram variants for that post are the exact ones excerpted above, produced in a single 3-minute run.