FILED FROM BANGALORE — The Web4 thesis the Bulletin has been tracking is, in the popular framing, a Bay-Area-and-Southeast-Asia story. The framing is incomplete. India is producing a third structural pattern — distinct from both — and Bangalore is the city where most of the pattern is visible. This piece is the working list of the five companies we think the field should be paying more attention to than it is.

The shape of the Indian autonomy-layer cohort matters because it is neither venture-pressured the way the Bay Area is nor patient-capital the way Southeast Asia is. It is government-adjacent, conglomerate-adjacent, and unusually willing to spend engineering effort on the parts of the autonomy stack the Western press has dismissed as solved. Foundation-model training. Sovereign infrastructure. Multilingual model alignment. None of these are the categories the Bulletin's directory has historically focused on. All of them are emerging as a recognizable Indian pattern.

Sarvam AI

Sarvam is the cleanest example of the pattern. The company, headquartered in Bangalore, is building a stack of Indic-language foundation models with an explicit thesis: that the next billion users of agentic systems will require models that are trained on, not retrofitted to, Indian languages. The technical work is serious — multilingual tokenizers built around the structural features of Indic scripts, training corpora curated against the under-representation problems Western open-source models have not addressed, evaluation infrastructure that respects the linguistic plurality of the country.

The commercial posture is the part that distinguishes Sarvam from a Western model-vendor. The company is positioned to serve the Indian government's emerging AI infrastructure mandate, the country's largest enterprises (which are mostly conglomerates rather than mid-market firms), and the developer base building agentic applications for Indian audiences. The Bulletin's read is that Sarvam is the working example of what a sovereign autonomy-layer foundation looks like.

Ola Krutrim

Krutrim is the Ola conglomerate's bid to be the Indian model-and-platform stack. The company has been more aggressive about public claims than Sarvam — talked about its own AI chip ambitions, talked about a full-stack vertical integration that includes hardware, model, and application — and the trade-press coverage has been correspondingly skeptical.

The Bulletin's editorial position is that the skepticism is partly fair and partly cosmetic. Krutrim is doing real engineering work; the model releases have been substantive, the platform integrations with the Ola ride-hailing and Ola financial stacks have been meaningful, and the company's hiring posture has attracted competent senior people. The hardware claims are harder to evaluate without more public data. We will track them.

What Krutrim's existence demonstrates is that the conglomerate-adjacent autonomy-layer pattern is real. India has a small number of large conglomerates with the patient capital, the engineering talent, and the strategic motivation to build full-stack autonomy infrastructure inside the corporate group. Reliance is likely to be next; Tata is rumored to be considering it. The structural conditions are unusual enough that they deserve their own editorial frame.

Devrev

Devrev is a more mature company than Sarvam or Krutrim — older, better-funded, and further into commercial deployment — but it belongs in this list because the company's posture has shifted meaningfully toward agentic operations over the last twelve months. The original Devrev pitch was about unifying customer feedback, product development, and support engineering inside a single tool. The 2025 evolution of the product is more straightforwardly an agentic operations platform for product and support teams.

The structural lesson is that the autonomy-layer transition is not only happening at the new-company layer. Established platforms are making the transition too, and Devrev is one of the cleaner working examples. The product is, by the time of this writing, a recognizable autonomy-layer product in a way that the 2023 version of the same product was not.

Bhanzu

Bhanzu — known earlier in the lifecycle as Exploring Infinities — is the directory's exception entry. The company is not strictly an autonomy-layer company; it is an EdTech company that has built an unusually opinionated agentic tutoring system underneath its consumer-facing product. We include it because the engineering work on the tutoring agent is, in our reading, more advanced than several of the products in the directory that explicitly claim to be autonomy-layer platforms.

The lesson is that the Indian autonomy-layer cohort is heavier on applied agentic systems than the Western framing tends to expect. The companies are building real products for real users, and the autonomy-layer engineering work happens underneath the product rather than as the product. This is the structural inverse of the Bay Area pattern, where the autonomy-layer engineering work is the product and the applied use case is left to the customer.

Lyzr

Lyzr is the working example of the Indian autonomy-layer agency posture. The company has built an opinionated framework for shipping agentic systems into enterprise environments and has staffed a delivery practice around it. The structural pattern is closer to the agency-platform pattern the Bulletin has tracked in Southeast Asia than to either of the other two Indian patterns described above.

We include Lyzr because the existence of an India-headquartered agency-platform firm is informative. It suggests that the agency-platform pattern is not regionally bound to any single ecosystem; it is a structural posture that emerges anywhere the commercial conditions reward a tight feedback loop between delivery and platform.

What the regional pattern is

The Indian autonomy-layer cohort, taken together, suggests three structural features the Bay Area pattern does not produce and the Southeast Asian pattern produces in different proportions.

The first is the conglomerate-adjacent capital structure. Indian autonomy-layer companies are unusually likely to be inside or close to a large corporate group. The patient capital is conglomerate, not venture, and the strategic horizon is correspondingly different.

The second is the sovereign-infrastructure orientation. Indian autonomy-layer companies are unusually likely to be building for, or with, the government's AI infrastructure mandate. The orientation produces a different set of priorities — multilingual coverage, on-device deployment, audit posture — than the consumer or enterprise-Western pattern.

The third is the applied-product orientation. Indian autonomy-layer companies are unusually likely to be shipping the agentic system as a feature inside a real product, rather than as the product itself.

The Bulletin's editorial position is that the Indian pattern is structurally important to the next phase of the autonomy-layer category, and that it has been under-covered relative to its actual contribution. We will continue to weight regional coverage toward Bangalore accordingly.