The Zeroic Journal

The AI desk

AI development · 16 pieces

How we build AI products that hold up in production: text-to-SQL, vector search, voice agents and the architecture around them. Written from systems we've built - Better Analyst (formerly FormulaBot) at 1.5M+ users, and voice, chat and data agents live with paying customers.

GPT-6.2 Sol vs Astra: which one your product should call

OpenAI shipped two new tiers on November 17. Which product features belong on each, what they cost per user, and what to change before you switch.

Prashant Abbi · 6 min read

Gemini 2.5 Flash deprecation: plan for the Dec 1 reroute

From December 1, Google answers your 2.5 Pro and 2.5 Flash calls with different models. What changes, what it costs in January, and how to pick a replacement.

Prashant Abbi · 8 min read

GPT-4o deprecation: move your app before January 14

What shuts down, what gpt-6-luna changes in your requests, what it costs, and a plan for fine-tuned models in the weeks left.

Prashant Abbi · 8 min read

OpenAI Agent Builder deprecation: your Nov 30 plan

What goes away on November 30, what stays, and a plan for the days left to move a support bot or internal agent into code you own.

Prashant Abbi · 8 min read

Claude Fable 5.2 pricing: when to pay 2x Opus 5.5

Anthropic's new top model costs 20% less than Fable 5.1 and still twice Opus 5.5. When a product should pay for it, and when a cheaper pattern does the job.

Prashant Abbi · 9 min read
Best tools

Best AI quiz generators for teachers and trainers (2026)

Nine AI quiz generators for schools, colleges and corporate training, grouped by market, and when building your own makes more sense.

Prashant Abbi · 9 tools

AI agent vs workflow: which one your product needs

Most products need a fixed outer flow with an agent inside one or two steps. Here is the test, the four limits to set, and a table to decide by task.

Prashant Abbi · 6 min read

How to reduce OpenAI API costs: five levers, ranked

Five ways to cut an OpenAI bill, priced on the same 300,000-request month and ordered by savings per day of work.

Prashant Abbi · 7 min read

Claude Haiku 5.5 cost: $0.10/M input tokens, a tenth of 4.5

Anthropic cut its small model's price tenfold on October 7. Here's what that does to your product's AI bill, and what to fix before you switch.

Prashant Abbi · 8 min read

Custom AI receptionist for clinics: build or buy

Most clinics should buy. Here are the three cases where building your own wins, what each option costs, and what a build involves.

Prashant Abbi · 6 min read

Jev vs LLM: when your app needs a decision model

TypeSafe's Jev picks from a fixed list and tells you how sure it is. Here's when that beats an LLM inside your product, and when it doesn't.

Prashant Abbi · 7 min read
Best tools

Best AI receptionists for clinics, hospitals and dental practices (2026)

Twelve AI receptionists for clinics and hospitals, market by market: what each one fits, what it costs, and when building your own makes sense.

Khushi Arora · 12 tools

GPT vs Claude vs Llama in production: choose by task type

Four kinds of LLM work, the model tier that wins each one, and the volume at which running open weights yourself starts to make sense.

Prashant Abbi · 8 min read

AI voice agent architecture: the four parts that decide if it ships

Where each layer of a production voice agent breaks on live phone calls, and the order we'd build them in, from the voice agents we've built.

Prashant Abbi · 10 min read

pgvector vs Pinecone: when Postgres stops being enough

The four signals that tell you pgvector has hit its limit, and the fixes to try inside Postgres before you pay for a second database.

Prashant Abbi · 8 min read

English-to-SQL at a million users: what breaks, in order

The five places every English-to-SQL pipeline breaks as it scales, from building FormulaBot's for 1.5M+ users, and the order to handle them in.

Prashant Abbi · 9 min read

Hear from us within 24 hours.

Prashant and Khushi Partners @ Zeroic

Stuck on something we’ve written about? The people who wrote these are the people who’d build it. Tell us what you’re working on - one of us replies within 24 hours.

Rather talk it through? Book a 30-minute call