Mailr — Wednesday, 21 October 2026
 |
| WED, 21 OCT, 2026 |
ISSUE # 87 |
|
|
Researchers got 0.8% of IndiaAI’s subsidised compute, while a handful of model builders got most of it.
Elsewhere, Amazon and US banks are turning AI agents away, Groww is building its own AI to cut token bills, and YouTube is labelling AI-made songs |
| |
|
📋 In today's Mailr:
|
|
|
🌍 Global AI |
|
Amazon and US banks shut out AI agents |
|
Image source: Google Nano Banana 2 |
| • |
Amazon blocked Meta's Muse agent in September; JPMorgan and Schwab ban agents from customer accounts entirely, citing password-sharing risks. Delta says no third-party agent partnerships exist and is still evaluating. |
| • |
Sites are building their own agent access instead of opening up. Cloudflare now offers default blocking and pay-per-crawl for AI bots, signalling that agent access will become a paid, permissioned layer. |
|
Bottom Line: Platforms are moving to monetise and control agent access rather than tolerate scraping. Any builder shipping AI agents that touch e-commerce, banking, or travel needs official integrations, not borrowed credentials. |
|
|
|
🇮🇳 India |
|
IndiaAI compute subsidy: researchers got 0.8% |
|
Image source: Google Nano Banana 2 |
| • |
Of ₹1,275 crore committed under the IndiaAI Mission, 83% went to 16 foundation-model approvals. Researchers got ₹10 crore (0.8%); government bodies like the CBI got nearly four times that in fully paid compute. |
| • |
Foundation-model bills were 98.9% subsidised; most others pay 60% upfront. One founder applying for under ₹10,000 in compute was asked to prove ₹50–200 lakh in revenue. He switched to AWS. |
|
Bottom Line: Public compute programmes that favour a handful of model builders over university labs undercut the research pipeline those same builders recruit from. Indian academics and early-stage founders should pressure IndiaAI for lighter paperwork and lower revenue thresholds. |
|
|
|
Groww builds in-house AI to cut token costs |
|
Image source: Google Nano Banana 2 |
| • |
Groww CEO Lalit Keshre says relying on third-party AI models would have cost "much higher", so the company is building its own AI stack. Cost to serve rose 24% quarter-on-quarter to ₹206 crore, partly attributed to AI and compute. |
| • |
Groww's GR-1 is an opt-in investing assistant that requires users to approve each action. Net profit fell 1% to ₹725 crore; operating revenue was flat at ₹1,502 crore. |
|
Bottom Line: Token costs, the per-use fees charged by AI model APIs, are becoming a real line item as AI usage scales inside consumer apps. Indian product teams running AI features at volume should model API costs against a build-your-own stack before locking into a provider. |
|
|
|
YouTube Music labels AI songs, but not audibly |
|
Image source: Google Nano Banana 2 |
| • |
YouTube Music now tags tracks as Fully Gen AI, Partly Gen AI, or No Gen AI using label declarations and its own detection. Labels don't affect monetisation; impact on recommendations is unconfirmed. |
| • |
India's IT Amendment Rules, in force since February, require synthetic audio to carry a prefixed audio disclosure and embedded metadata, not just a visual tag. It's unclear whether YouTube's approach satisfies them. |
|
Bottom Line: India's IT Amendment Rules require synthetic audio disclosures inside the audio itself, not just on screen. Builders shipping AI-generated audio for Indian users must plan for spoken disclaimers and embedded metadata, or risk non-compliance. |
|
|
|
⚡ Mailr Signal |
|
|
The IndiaAI compute data reveals a structural contradiction: the mission subsidises model builders heavily but prices out the university labs that train the engineers those builders hire. If the pattern holds, India risks building a foundation-model layer with a hollowed-out research base beneath it.
Our take, not a news summary.
|
|
|
🛠️ Tool of the Day |
|
|
skit.ai
GenAI-powered omnichannel voice, SMS, email, chat assistants
| 🇮🇳 |
Founded in Bengaluru, India; Cyllid Technologies reported 10-50 crore rupees annual revenue as of March 2025. |
| ✅ |
Named an IDC Innovator in Voice AI (April 30, 2025). |
| 💡 |
Best for: Debt collection teams automating high-volume outreach with compliance and empathy across channels. |
View tool →
|
|
|
📱 Social Scan |
|
 Asked for the ultimate life hack, Gemini recommended the Hemingway Bridge: stop working when you're on a roll, not when you're done. Leave yourself an obvious next step so restarting tomorrow is easier. Apparently, the secret to productivity is quitting at the right time.
View post →
|
|
|
📖 Read of the Week |
|
|
Long-running agents unlock outcome-based pricing
by David Haber
| • |
AI agents now handle multi-week customer journeys end-to-end, replacing fragmented handoffs. This shift enables outcome-based pricing—charging on closed deals rather than API calls—moving agents from cost-cutting tools to revenue-sharing partnerships across mortgages, insurance claims, and enterprise sales. |
| • |
Map your full customer journey and design agents to own the entire relay race, not single tasks. Reprice from per-call to per-outcome models. Target sectors with multi-party workflows: MSME lending, insurance claims, legal intake, real-estate transactions where long-horizon reliability creates defensible business. |
|
|
|
|
ପୁଣି ଦେଖା ହେବ
Thanks for reading. If something felt off or overhyped, reply and call it out — your feedback makes every edition sharper.
The next edition is already taking shape. Don't miss it.
Were you forwarded this? Join free at mailr.tech
|
|
|
You're receiving this because you subscribed to Mailr.
© 2026 Mailr
|
|
|