Article URL: https://www.economist.com/by-invitation/2026/08/13/the-case-for-overhauling-american-science Comments URL: https://news.ycombinator.com/item?id=49305708 Points: 4 # Comments: 0
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
A developer reviewed every major open-source WhatsApp bot framework on GitHub and concluded they all share a fundamental architectural flaw: they function as basic API wrappers that add generic AI without domain-specific capabilities. Existing tools route messages and append conversation history to a language model, but lack the industry-specific logic needed for businesses like restaurants or law firms that require distinct workflows and data access. Enterprise platforms from providers such as Twilio and Intercom charge up to $2,000 per month but similarly fail to integrate meaningfully with a client's own systems or data. The developer also highlighted that most frameworks ignore critical production concerns such as GDPR-sensitive PII flowing through third-party APIs and the risk of service outages from a single AI provider. In response, they built SARA, an open-source WhatsApp AI agent platform featuring 20 vertical-specific agent profiles, each equipped with its own tailored toolset and function-calling capabilities.
Leonardo de Moura, creator of the Lean proof assistant, was interviewed about how formal verification can guarantee the absence of software bugs — going beyond what traditional testing allows. Lean functions both as a programming language and a proof system built on dependent type theory, enabling mathematical and software verification within a single platform. The Lean nonprofit foundation, established in 2023 with major support from AWS, is now accelerating Lean's role as a programmable software verification system after a decade as an academic project. De Moura highlighted a key future use case: AI-generated code optimizations paired with machine-checked proofs of behavioral equivalence, removing the risk from automated refactoring. He also recommended learning Lean today by pairing it with AI agents, using official resources such as 'Theorem Proving in Lean' and 'Functional Programming in Lean'.
A developer has published a lightweight Python script designed to track and limit token consumption when using free AI model endpoints. The tool maintains a local JSON ledger that estimates token costs before each API call and records actual usage afterward, refusing to proceed if the projected total would exceed a set budget. The approach addresses common pitfalls such as runaway retry loops and oversized context buffers that can silently drain a free tier's token allowance before dashboard alerts trigger. The script is built around the OpenAI-style chat completion format and can be adapted to other compatible endpoints by swapping a single function. The article was written as part of promotional outreach for MonkeyCode, an open-source project offering a free model route and hosted server.
Developers using free AI model tiers often discover quota overruns only after a batch job fails, wasting both time and retry budget. A lightweight Python token ledger can predict whether a batch job fits within a given allowance before any live API call is made. The approach uses a rough heuristic of one token per four characters to estimate prompt and completion costs across thousands of calls. For example, a job requiring 6,000 summaries with 21,000-character prompts would need an estimated 32.25 million tokens, exceeding a 30-million-token free allowance. The ledger does not replace a real tokenizer but serves as a fast, deterministic pass/fail filter that separates budget planning from actual API spending.
Discussion (0)
Log in to join the discussion and vote.
Log in