We hosted a Foundry Live Demo Night on October 13th at Microsoft in San Francisco.

It was a great night to see the latest applications of Agentic AI from Bay Area startups.

The presentations covered marketing, agentic APIs and MCP, space/geo-spatial AI, and personalized AI.

The evening began with an overview of ways startups can partner with Microsoft, presented by Minsoo Lee on the Corporate Business Development team.

Microsoft Foundry is Microsoft’s AI platform that helps startups quickly build, deploy, and scale AI applications and agents using leading models, enterprise-grade infrastructure, and built-in security and governance.

Learn more about the startups and watch the demo video below.

Presenters

Hightouch

Hightouch, presented by Alex Tirion, is the creator of Ad Studio, an AI marketing platform that enables marketing teams to create, edit, and launch on-brand ads across advertising platforms — incorporating brand guidelines, business context, performance goals, and human guidance. It helps differentiate campaigns and reduce ad fatigue.

Liner

Liner, presented by Luke Kim, is an AI platform for professionals that combines cited web search, academic research, and writing tools to help users find information and create polished documents. They offer an enterprise API for model routing and optimization. They recently launched Liner Actions MCP, a unified connector that lets agents interact with 1,000 verified applications via a single setup.

Orbit AI

Orbit AI, presented by Gus Liu, is building a distributed AI cloud spanning Earth and space, combining terrestrial GPUs with satellite-based compute to run AI inference and process Earth-observation data directly in orbit. They currently have three satellites in orbit.

Uare.ai

Uare.ai, presented by Sam Horton, is an AI platform that lets individuals build, own, and control a personalized AI trained on their voice, knowledge, experiences, and way of thinking, which they can use to create content, interact with others, and share their expertise. One example is that of a 28-year career math educator who works with 20,000 students and 54 school boards, who was able to scale and sell his curriculum and courses through the personalized AI that students can interact with, in a personalized fashion.

Watch the video

Next event

Our Next event is November 10th at the GitHub Headquarters in San Francisco. Register to attend.

Arte Merritt is a Director of Business Development and Strategic Partnerships, AI Startups at Microsoft, where he focuses on strategic partnerships and go-to-market with leading AI startups. Previously, at AWS he drove startup adoption of Bedrock and AgentCore, and created and led AWS’s Global Conversational AI Partner program. He is a serial entrepreneur having founded two, tier-1 VC-backed startups, one of which he sold to Nokia, and the other he grew to 20,000 customers and received six acquisition offers. He is a frequent author and speaker on Generative AI, Conversational AI, and data insights. Arte is an MIT alum.

Originally published at https://www.linkedin.com.

We hosted a Foundry Live Demo Night on September 15th at the GitHub headquarters in San Francisco.

It was a great night to see the latest applications of Agentic AI from Bay Area startups.

The presentations covered computer use agents, security, GTM automation, logistics, and file storage.

Microsoft Foundry is Microsoft’s AI platform that helps startups quickly build, deploy, and scale AI applications and agents using leading models, enterprise-grade infrastructure, and built-in security and governance.

Learn more about the startups and watch the demo video below.

Presenters

Cua

Cua, presented by Dillon DuPont, is a platform for building, running, training, and evaluating computer-use agents (cua) across browsers and Linux, Windows, MacOS, and Android environments. Cua Driver is the core component — an MCP for agents that gives background computer use on any platform. Cua Spaces provides a sandbox in the cloud. It enables users to drag, or “teleport,” local applications like chrome into Spaces to continue running in the cloud.

Konvu

Konvu, presented by Lucas Masson, is an agentic security platform that investigates vulnerabilities flagged by existing scanners, determines which are exploitable, and automatically generates fixes and pull requests. To prevent issues even earlier, Konvu created a guardrails product that gives agents the right security context to help prevent agents from introducing vulnerabilities in the first place.

Kris at Work

Kris at Work, presented by Chirag Hiran, is an agentic GTM platform that helps revenue teams automate and optimize prospecting, deal execution, customer retention, and CRM management. It helps find the right customer, at the right time, with the right context. The platform provides signals and recommendations for the sales approach, along with hyper-personalized messaging given the context of the seller as well as the prospect.

Oway

Oway, presented by Phillip Nadjafov, is an agentic logistics platform for supply chain and industrial manufacturing. They created the Internet of Industrials (“IOI”) — a standardized context layer that allows all the different legacy data systems, robots, vehicles, factories, and more to interact with each other. One of their product, Juno, tracks freight space and vehicle location to optimize matching shipments with unused capacity, in real time.

Space

Space, presented by Matt Ao and Ari Bapna, is a cloud-based filesystem, with infinite space, that appears local, enabling users to interact with large files while taking up very little local disk space. For example, a 200MB video file will show up as ~0 bytes locally but will immediately open and play, and be available for use in applications. The service can also be used by agents to quickly hydrate sandboxes, simply by mounting the Space filesystem in the sandbox.

Watch the video

Next Event

Our Next event is October 13th at the Microsoft Reactor in San Francisco. Register to attend.

Arte Merritt is a Director of Business Development and Strategic Partnerships, AI Startups at Microsoft, where he focuses on strategic partnerships and go-to-market with leading AI startups. Previously, at AWS he drove startup adoption of Bedrock and AgentCore, and created and led AWS’s Global Conversational AI Partner program. He is a serial entrepreneur having founded two, tier-1 VC-backed startups, one of which he sold to Nokia, and the other he grew to 20,000 customers and received six acquisition offers. He is a frequent author and speaker on Generative AI, Conversational AI, and data insights. Arte is an MIT alum.

Originally published at https://www.linkedin.com.

We hosted a Foundry Live Demo Night on August 18th at the GitHub headquarters in San Francisco.

It was a great night to see the latest applications of Agentic AI from Bay Area startups.

The presentations covered agentic GTM automation, marketing content workflows, customer service operations, model routing, governance, and fine tuning.

The evening started with an introduction to Microsoft for Startups by Heena Purohit, the Director of AI for Microsoft for Startups.

Microsoft Foundry is Microsoft’s AI platform that helps startups quickly build, deploy, and scale AI applications and agents using leading models, enterprise-grade infrastructure, and built-in security and governance.

Learn more about the startups and watch the demo video below.

Presenters

4Minds

4Minds, presented by Kevin Locke, is a contextual AI platform for fine-tuning, deploying, and continuously improving custom language models, trained on an enterprises data. Users can give the model a persona and select the data to ingest into a knowledge graph, through 40 out-of-the-box integrations, to create the context to fine-tune the model.

Gradial

Gradial, presented by Connor Davis, is an AI marketing-execution platform that automates marketing operations and content workflows from strategy ideation through deployment. The platform integrates with CMS, brand management, and work tracking systems to act as a glue across them. An example use case is having Gradial take a Figma design and build out all the pages in a CMS with the extracted assets.

HappyRobot

HappyRobot, presented by Brooke Adair and Erena Estelles, is an agentic operations platform that deploys autonomous voice and workflow agents to handle communications, decisions, coordination, and complex processes in customer service, logistics, and supply-chain workflows. For example, in their utility company demo, the AI agent took the customer’s call, understood the issue, applied discounts when the customer expressed frustration, and then figured out the closest technician and called them with all the details to coordinate an appointment, and finally returned to the customer call to let them know when the technician would be there.

Mesh API

Mesh API, presented by Raushan Sharma, is a model routing service that provides access to over 1,000 models, along with additional features for agentic development, including MCP, persistent memory, and RAG. The platform features a governance layer to control model access and usage by organization, team, and individual, along with an audit trail. There is a visibility layer into token usage, spend, and even errors.

Sprouts

Sprouts, presented by Karan Chaudhry, is an agentic GTM platform that helps sales and marketing teams identify prospects, enrich account data, and automate outreach to drive growth and improve conversion rates. For example, Sprout can research and identify prospects based on natural language and dive deeper into org charts to find mutual connections for warm intros. The platform can be integrated with CRMs via MCP.

TrueFoundry

TrueFoundry, presented by Rishikesh Anand, is an AI platform for securely deploying, governing, and monitoring models, agents, and applications across cloud and on-premise infrastructure. The platform enables customers to create custom rule sets for how they would like to route across models.There is also a skills registry to maintain reusable combinations of context, prompts, and logic across agents.

Watch the video

Next event

Our Next event is September 15th at the GitHub Headquarters in San Francisco. Register to attend.

Arte Merritt is a Director of Business Development and Strategic Partnerships, AI Startups at Microsoft, where he focuses on strategic partnerships and go-to-market with leading AI startups. Previously, at AWS he drove startup adoption of Bedrock and AgentCore, and created and led AWS’s Global Conversational AI Partner program. He is a serial entrepreneur having founded two, tier-1 VC-backed startups, one of which he sold to Nokia, and the other he grew to 20,000 customers and received six acquisition offers. He is a frequent author and speaker on Generative AI, Conversational AI, and data insights. Arte is an MIT alum.

We hosted a Foundry Live Demo Night on July 21st at the GitHub headquarters in San Francisco.

It was a great night to see the latest applications of Agentic AI from Bay Area startups.

The presentations covered agentic commerce, agent building, insights into agentic efficiency, and agentic web services.

Microsoft Foundry is Microsoft’s AI platform that helps startups quickly build, deploy, and scale AI applications and agents using leading models, enterprise-grade infrastructure, and built-in security and governance.

Learn more about the startups and watch the demo video below.

Presenters

Crew AI

CrewAI, presented by Lorenze Jay Hernandez, is a platform and open-source framework for building, orchestrating, and deploying collaborative AI agents that automate complex workflows. CrewAI flows and orchestration layer enable agentic systems to balance both deterministic outcomes and agency.

Faros

Faros, presented by Chase Norton, is an engineering intelligence platform that unifies data from across the software development lifecycle to give engineering leaders visibility into software delivery, productivity, quality, and operational performance — especially AI spend and outcomes. Faros provides insights into token spend, efficiency, and impact on results.

SkyFire

SkyFire, presented by Craig DeWitt and Ankit Agarwal, is an agentic identity and commerce platform that enables autonomous agents to securely authenticate, interact, and transact with online services. Their know your agent (KYA) token enables agents to interact with websites through the “front door” like a human, and not be blocked like a bot. They currently integrate with 70% of the commercial web.

Sycamore

Sycamore, presented by Sri Viswanath, is an agentic platform for building, deploying, and orchestrating agents — an “agentic operating system.” One of their products, Forge, is an interface that enables users to build agents and automated workflows, via natural language. Forge can be used by anyone in the company — both technical and non-technical users. It can leverage organization knowledge and data, and continuously learns over time.

TinyFish

TinyFish, presented by Mark Peng, is a service that enables autonomous agents to have access to the web — for web search, browser control, and web automation. The web search is optimized for speed and reliability from an agent perspective. Their fetch service is optimized for relevant information density — able to get the relevant context and data while ignoring ads and irrelevant info. The browser control API is able to leverage browser context profiles to restore sessions and manage identity.

Watch the demos

Next event

Our Next event is August 18th at the GitHub Headquarters in San Francisco. Register to attend.

Arte Merritt is a Director of Business Development and Strategic Partnerships, AI Startups at Microsoft, where he focuses on strategic partnerships and go-to-market with leading AI startups. Previously, at AWS he drove startup adoption of Bedrock and AgentCore, and created and led AWS’s Global Conversational AI Partner program. He is a serial entrepreneur having founded two, tier-1 VC-backed startups, one of which he sold to Nokia, and the other he grew to 20,000 customers and received six acquisition offers. He is a frequent author and speaker on Generative AI, Conversational AI, and data insights. Arte is an MIT alum.

Originally published at https://www.linkedin.com.

We hosted a Foundry Live Demo Night on June 16th at the GitHub headquarters in San Francisco.

It was a great night to see the latest applications of Agentic AI from Bay Area startups.

The presentations covered Agentic AI super agents, incident resolution, browser control, and manufacturing simulations.

Microsoft Foundry is Microsoft’s AI platform that helps startups quickly build, deploy, and scale AI applications and agents using leading models, enterprise-grade infrastructure, and built-in security and governance.

Learn more about the startups and watch the demo video below.

Presenters

DevRev

DevRev, presented by Ahmed Bashir and Eric Gan, is the maker of Computer — a conversational AI agent that leverages shared memory from multiple data sources along with organizational workflows and context to enable employees to get answers, automate work, and take actions across support, sales, operations, and more.

Neubird

Neubird, presented by Andrew Lee, is a “production operations agent” — an AI SRE Agent that autonomously investigates incidents, identifies root causes, and recommends and executes remediation. Andrew showed a comparison of using Claude Code to “operate” on code versus Neubird running an investigation — highlighting the benefits of Neubird including the ease of use, deeper insights from shared context and knowledge, and prioritized actions to take.

NinjaTech

NinjaTech, presented by Babak Pahlavan, is an Agentic AI platform featuring autonomous AI “employees” and long-running “super agents” capable of collaborating with humans, using tools, writing code, conducting research, and completing complex workflows across enterprise applications. The agents are self-learning and self-evolving. Users can interact with them in Slack and Teams much like interacting with a human, and the agents can interact with each other too.

SolverX

SolverX, presented by Andy Junso Park, is a “physics intelligence” AI platform that performs simulations and analysis to predict results that can solve underlying engineering problems in manufacturing in seconds rather than hours or days. For example, the platform can perform a thermal buckling simulation on a car hood or the stress contour on a car door.

Yutori

Yutori, presented by Eric Tay, is a web agent platform and the maker of Navigator N1.5 — a computer-use model that autonomously interacts with web browsers much like a human would — i.e. clinking, typing, scrolling, navigating, and executing multiple tasks. The model is available via an API and MCP and compares favorably well on accuracy, latency, and cost versus Anthropic and OpenAI.

Watch the video

Next Event

Our Next event is July 21st at the GitHub Headquarters in San Francisco. Register to attend.

Arte Merritt is a Director of Business Development, AI at Microsoft, where he focuses on strategic partnerships and go-to-market with leading AI startups. Previously, at AWS he drove startup adoption of Bedrock and AgentCore, and created and led AWS’s Global Conversational AI Partner program. He is a serial entrepreneur having founded two, tier-1 VC-backed startups, one of which he sold to Nokia, and the other he grew to 20,000 customers and received six acquisition offers. He is a frequent author and speaker on Generative AI, Conversational AI, and data insights. Arte is an MIT alum.

Originally published at https://www.linkedin.com.

Agents are rewriting how we build, invest, and work.

On Tuesday, I had the pleasure to moderate a fascinating panel at the GitHub office in San Francisco, with three leaders at the forefront of this agentic shift: Ash Lewis, CEO and Co-founder of Fastino Labs; Cheryl Cheng, Managing Partner at Microsoft’s M12 Ventures; and Marko Kraemer, CEO and Founder of Kortix.

Our discussion dove into where agentic AI is delivering real impact today — from engineering processes to investor due diligence to how we interact with these models. We also explored where there are opportunities for startups to differentiate and create moats in this highly competitive space.

The video recording is at the end

What makes something Agentic?

We hear a lot about Agentic AI. There is a lot of buzz around “agents,” but what makes something truly agentic versus other Generative AI applications, chatbots, or RPA?

Marko offered the following description: from an LLM perspective, instead of having an LLM that you chat with back and forth, you give an LLM an environment, let it take actions, feed the results back in, and let it decide what to do next — with increasing autonomy.

Ash expanded on this, that it is not just an LLM with a series of tool calls. A truly agentic system can self correct, and maybe even write its own tools.

The perception-action-feedback loop is the through-line.

When it comes to deciding when to build an agent, Marko provided a useful framework for picking agentic automation targets. The best tasks today are the ones with vast input and output data, and a fairly-well defined standard operating procedure (SOP). For example, recruiting sourcing, insurance claims processing, or anything where you apply intelligence within a standardized procedure.

Open-ended tasks can work too, but need much more human judgment in the loop. The “go Claude, make me $10 million” meme falls apart for obvious reasons. However “spin up an outbound marketing agent to help me hit $1M” is a possibility.

Where is the excitement today?

The panelists shared what gets them excited about Agentic AI.

Cheryl framed the agentic AI opportunity as two bookends. At one end is the work humans don’t want to do — i.e. the mind-numbing stuff like reconciling checks against ledgers. At the other end, is work humans can’t do — like compressing weeks or years of drug-discovery research into something tangible. Both ends carry a lot of economic rent, and they have very different implications for the workforce.

Ash’s excitement was more personal. He found running a company is genuinely a lot of fun when there are three to five people, and he is looking forward to running companies that small again — with agents filling in the gaps. He also pointed out how agents do not suffer from fatigue.

Marko tied it back to AGI. The motivator for him with agents is the same: we are automating intelligence, which enables us to work faster, smarter, and harder on a scale we never could before. What a small team, or solo founder is able to accomplish with agents today is incredible.

The impact on software development

Agentic AI is having a significant impact on the development process.

At Kortix, the biggest application of Agentic AI internally is engineering. They are using AI to build more AI. As Marko puts it, if you have an engineering team that is not using a coding agent today, you are very far behind. His team is trying to squeeze the most out of every token and computer — looking to see how they can have their agents keep working while they sleep.

They re-engineered their observability stack so that when the same error happens enough times, an agent automatically goes out resolves the problem. They do the same for PR reviews, which matters when you are an open-source project receiving tons of PRs. They have extended the same playbook into marketing — creating a “social engine” agent that monitors X and LinkedIn for specific terms and posts and interacts when there is a mention.

Ash had a great observation about the impacts on the development process. He is used Agile for a decade, but for the first time, it is not working because his team can move so fast. They would do sprint planning, and then all 50 tickets would be done two days in, way before the sprint is over. It has also led to some unintended consequences where engineers are stepping on each other’s toes because everyone can make things so fast.

For Ash, agents go far beyond development in his daily work. He created a project he calls Rocky, that is essentially a Kanban board for managing Claude instances. Everything from booking flights to production code changes flows through it. Almost every email, flag, and task he handles now has an LLM involved — which he points out is useful, but “a little creepy,” given how much cognitive load he has offloaded.

To CLI or not to CLI? That is the question…

An interesting debate arose amongst the panel on command line interfaces versus application user interfaces — how builders interact with agents versus how everyone else does.

When Fastino launched their product, Pioneer, they had spent two months obsessing over the UI design. Yet, the first question they got after launch was, “where is the CLI?” — and the most viewed page on the site is the docs. Ash questioned whether they may have wasted time building a UI.

Marko also recommended using the CLIs of products as a cool hack. He disliked using the console interfaces of cloud infrastructure providers. Then he switched to using the CLIs and loved it, as the CLI agent “just understands and gets it.” He no longer looks at dashboards or even his own code anymore.

Cheryl, however, pushed back on CLIs — as a self-described non-technical person, she would never live in a CLI. She does not want a blank screen and “build yourself an agent.” She wants a simple interface, and, interestingly, she does not want as much choice as builders assume she does. She pointed out that those of us working in agentic AI are at the tip of the distribution and adoption curve, whereas the rest of the world is not. There are folks who enjoy reading and processing things on their own.

She also flagged a real concern — agentic AI compresses tasks down to a small amount of time so well, that it can trade away the last 20% of time that actually produces creative breakthroughs. The boredom of grinding through a problem is sometimes what forces you to think differently. Marko agreed — there is a real risk that high-velocity AI workflows feel like brain rot, and an open question whether faster token speeds amplify creativity or kill it further.

In addition to CLIs and web interfaces, voice is also a popular option. Apparently if you speak your prompts, the added context makes the results better. Ash described a YC founder who walks around with a lapel mic, narrating prompts to an agent throughout the day. Marko seconded this as he dictates everything into Whisper and runs up to five parallel threads at once.

The impact on VC fundraising

As agents are making it easier for smaller teams to do more, I asked Cheryl whether she is seeing more solopreneurs raise venture money. While she does see it, at M12, as a strategic investor, most of the companies they see have much bigger teams than that. M12 looks at where they can be strategic to the enterprise and knowledge worker. It is rarer to find solopreneurs focusing on the enterprise — they tend to be more on the consumer side.

What she is seeing though, is very small teams doing a lot, which is changing how her team does due diligence. She now asks founders not just about the product, but about the internal tools they are building for their own team. The internal stack is becoming almost as important as the product stack, because that is where compounded leverage with AI actually shows up.

The risk profile has shifted too. The floor keeps dropping as CFOs and CTOs increasingly say, “my friend, Claude, and three engineers could build this.” Some VCs are now having junior staff spend a weekend with Claude trying to recreate what they just heard pitched — and if a meaningful chunk of it gets built in a weekend, that is a real signal about defensibility. That being said, Cheryl pointed out that the “choke point” in enterprise is still distribution — and that may be where one could have a moat.

Where are the moats?

With all the agentic tools from code gen to marketing and sales, making it easier to build, along with frontier models adding new functionality seemingly daily that competes directly with whole categories, where are the opportunities for a startup to differentiate?

In addition to distribution Cheryl mentioned earlier, she also sees an opportunity at the application layer where the work actually happens and a lot of value lives.

One area of opportunity is in the unglamorous, essential workflows that internal enterprise developers do not want to build — ones that are not their idea of a good time — like an accounts receivable or sales tax application for the office of the CFO. Selling into large enterprises with big developer orgs is hard right now, as those teams want to build themselves, and they often have legitimate IP reasons for doing so. However these unglamorous, but important workflows present an opportunity to build a moat.

The other area is with mid-market firms who do not have enough resources to build applications on their own. They are more willing to buy.

The important piece is for the startup to learn the workflow in a process and improve over time with data, to be entrenched. Then they will not be in the existential threat of an internal enterprise developer saying they are going to use Claude and try and build it on their own.

Cheryl brought up an interesting point, and another potential area of opportunity — taking a different approach with agents. A lot of startups are building agentic solutions that mirror human processes — just doing them more or faster. But what if humans were not there — how would the process have been designed with just agents in mind? Folks are rushing to build agents, as opposed to rethinking the processes before they build an agent.

As a founder, Marko has spent a lot of time thinking about potential moats in this space. A lot of companies appear very similar — for example: Manus, Genspark, Lovable, OpenCode, Claude Code, Perplexity, and Bolt.new. From a technology standpoint they are all pretty much the same — with an agents paradigm and skills paradigm. Where they actually differentiate though and have a moat is in the operational knowledge and data accumulated over time. There are also aspects of the product as well — the cognitive architecture (how memory is handled), the skills and workflows, and the actual UI/UX. For example, Lovable, he argued, is still a better web-app builder than newer competitors because it is purpose-built and has a better user experience with one-click integrations to GitHub, the cloud, databases, and more.

Predictions for the future

Each of the panelists provided thoughts on the future of Agentic AI.

Ash believes we will start to see more proactivity in agents. He told a funny story about receiving a heart-shaped rock in the mail — briefly believing he had a secret admirer, only to realize his agent, which had access to his credit card, bought it for him. Funny story aside, there is a lot of potential for agents being more proactive. Once an agent has access to your signals — email, messages, calendar — it should not need to be asked to book a flight or the hotel. It should already know and take action.

Cheryl predicts a sharpening of ROI measurement of agentic AI. Are the tokens that we are burning through worth it? Organizations are still experimenting, but in the near future we will need FinOps-like tools for tying outcomes back to whether something should have been an agentic workflow or not. This is especially important as we will most likely continue to be in a capacity constrained world for a while.

Marko feels we will see more autonomous agents and companies — heading to a roughly 80/20 split of agents to humans. Intelligence is getting automated. Humans will become high-level orchestrators where our taste ability to judge is what matters. He cited a paper that argues the main blocker to scaling AGI is human verification capability. Even now, with intelligence more abundant than ever, it comes down to hiring the best people — so your verification capabilities stay sharp.

Conclusion

It was great having an opportunity to learn more from Ash, Cheryl, and Marko. The agentic shift is really about redesigning work, teams, and the interfaces we use, around a technology that does not get tired and never stops iterating.

I am excited to see what the future has in store.

Watch the video

Arte Merritt is a Director of Business Development, AI at Microsoft, where he focuses on strategic partnerships and go-to-market with leading AI startups. Previously, at AWS he drove startup adoption of Bedrock and Agentic AI, and created and led AWS’s Global Conversational AI Partner program. He is a serial entrepreneur having founded two, tier-1 VC-backed startups, one of which he sold to Nokia, and the other he grew to 20,000 customers and received six acquisition offers. He is a frequent author and speaker on Generative AI, Conversational AI, and data insights. Arte is an MIT alum.

Originally published at https://www.linkedin.com.

We hosted our second Built on Bedrock demo night for 2026 on Feb 17th at the AWS Builder Loft in San Francisco.

It was a great night to see the latest applications of Agentic AI from Bay Area startups.

The presentations covered agentic AI observability, incident resolution, and clinical trial administration.

Each of the startups who presented works with Amazon Bedrock — Amazon’s fully managed generative AI service.

The evening started with a demo of the Amazon Bedrock AgentCore CLI by Jean Malha. The AgentCore CLI is a command line interface that makes it easier to build and deploy Agentic AI on AWS using AgentCore.

Learn more about the startups and watch the demo video below.

Presenters

Cleric

Cleric presented by Shahram Anver is an agentic SRE that investigates issues in production, alerts of the issues, and provides suggestions on how to fix. The agent is able to learn from previous interactions to better handle future issues.

Firetiger

Firetiger presented by Achille Roussel is an agentic, observability platform that identifies and fixes problems in production through observability data, code base understanding, and knowledge of the business. FireTiger launched their platform publicly on Feb 18th, and announced a $7.6M seed round.

Grafana Labs

Grafana presented by Rob Weidner and David Tupper is a full-stack, observability platform that can identify issues and propose fixes. Grafana Cloud unifies metrics, logs, traces, and dashboards in one place for a complete monitoring solution for applications, infrastructure, and services. Users can interact via a chatbot assistant via natural language to investigate issues or better understand the environment.

The Contract Network

The Contract Network (TCN) presented by Phil Richards is an agentic platform to help with clinical trials. It can take anywhere from 90 to 180 from the time drugs are developed to when a pharmaceutical company can start clinical trials. Most of the time is on administrative tasks dealing with contracts, budgets, lab manuals, and similar things. TCN is able to reduce that time through its agentic platform — they started with contracts and can now do the full administration of the clinical trial.

Watch the video

Next Event

Our next event is March 10th at the AWS Builder Loft in SF. Stay tuned for more info on registering.

If you are interested in presenting, fill out the demo application.

Arte Merritt is a GTM BD for Agentic AI for startups. He is the founder of Reconify, an analytics and optimization platform for Generative AI. Previously, he led the Global Conversational AI partner initiative at AWS. He was the founder and CEO of the leading analytics platform for conversational AI, leading the company to 20,000 customers, 90B messages processed, and multiple acquisition offers. He is a frequent author and speaker on Gen AI, Conversational AI, and data insights. Arte is an MIT alum.

We hosted our first Built on Bedrock demo night for 2026 on Tuesday at the AWS Builder Loft in San Francisco.

It was a great night to see the latest applications of Agentic AI from Bay Area startups.

The presentations covered agentic AI development, security, and analytics, as well as use cases in image analysis and healthcare and life science research.

Each of the startups who presented works with Amazon Bedrock — Amazon’s fully managed generative AI service.

The evening started with a demo of the Amazon Bedrock AgentCore CLI by Jean Malha. The AgentCore CLI is a command line interface that makes it easier to build and deploy Agentic AI on AWS using AgentCore.

Learn more about the startups and watch the demo video below.

Presenters

Bolt.new

Bolt.new, presented by William Sayer, is a vibe coding platform for creating websites and applications. William demonstrated creating a website via simple prompt as well as a more sophisticated application via a product requirements doc (PRD) that was also created via prompts. They primarily focus on B2B and incorporate security and compliance into the applications. Their revenue grew from $700k ARR to $40M in five months.

Britive

Britive, presented by Shahzad Ali, is an identity security SaaS platform. Britive offers solutions to prevent security breaches. They provide “just in time” access — short lived, scoped access to resources when a user needs it, and then remove it afterwards. It works with agentic AI agents as well as human access. Britive works with AgentCore and Bedrock.

CoreViz

CoreViz, presented by Wassim Gharbi, is a workspace for teams and organizations that work with thousands of photos. The platform creates descriptions and indexes photos to enable search and object detection. Users can modify collections of photos too — for example changing the background of a group of wedding photos all at once. They are seeing traction in forensics and medical use cases — shoe print analysis and matching, xray diagnostics, and more.

Distributional

Distributional, presented by Scott Clark, is an agent analytics platform. The platform analyses all of the logs being generated by Agentic AI agents to help customers understand behavior, identify patterns, and find areas to fix or improve. It supports OTEL to ingest data. They can provide topic and behavior analysis based on clusters and patterns. Distributional helps enterprises identify the “unknown unknowns” and offers suggestions on possible fixes.

Manifold

Manifold, presented by Rajendra Koppula, is a data and AI platform, purpose-built for life sciences and healthcare. The platform helps companies organize data including genomics data, imaging data, and other operational health data. It then enables researchers to do analysis on the data and collaborate with others.

Watch the demos

Next Event

Our next event is February 17th at the AWS Builder Loft in SF. Register to attend.

If you are interested in presenting, fill out the demo application.

Arte Merritt is GTM specialist for Agentic AI for startups. He is the founder of Reconify, an analytics and optimization platform for Generative AI. Previously, he led the Global Conversational AI partner initiative at AWS. He was the founder and CEO of the leading analytics platform for conversational AI, leading the company to 20,000 customers, 90B messages processed, and multiple acquisition offers. He is a frequent author and speaker on Gen AI, Conversational AI, and data insights. Arte is an MIT alum.

Trends and Predictions from the experts

This past year has been an amazing time for Generative AI. The continuous advancements in models and applications built on top of them, has been at such a rapid pace, it is reminiscent of the boom and excitement of the early Internet days.

I was fortunate to have the privilege to moderate a panel of experts, each with a unique vantage point into startups leveraging Generative AI: Anamitra Banerji, co-founder and Managing Partner of Afore Capital; Barath Chari, Partner at Wilson Sonsini; and Sam Lisowski, Senior Sales Leader on AWS’s Gen AI startups team.

The conversation revealed not just where we are in the Gen AI journey, but where we are headed. Here are the key insights that emerged from our discussion. The video recording is at the end.

The great AI migration: talent floods back to San Francisco

One of the more interesting observations came from Anamitra, who noted a massive influx of young founders and entrepreneurs to San Francisco this year. “It’s reminiscent of 2010,” he explained, “back when the pendulum finally swung from Menlo Park to San Francisco with companies like Twitter, Dropbox, and Uber.” After the pandemic exodus, Gen AI has become the magnetic force drawing talent back to the city. We especially see this with early stage startups enrolling in YC and moving to the DogPatch during their cohort.

This migration is not just about geography. It represents a fundamental shift in entrepreneurial energy. Young technologists, fresh out of school, are coming to San Francisco specifically because of Gen AI opportunities. The volume of startups Afore Capital is seeing has increased radically, signaling that we are in the midst of a genuine entrepreneurial boom.

The sheer funding percentage concentrated in the Bay Area is astounding. Crunchbase estimates ~71% of global funding was to startups based in the US, and ~70% of that was to startups in the SF Bay Area.

From benchmarks to business: the shift to real applications

Perhaps the most significant trend the panel identified is the maturation of the AI ecosystem. As Anamitra put it, “The conversation has moved on from benchmarks or improvements in foundation models to actual applications, to actual deployments in the enterprise.”

This shift from theoretical capabilities to practical implementation is profound. We are no longer talking about prompts and prompt engineering in isolation. Instead, the focus has turned to Agentic AI solutions that do actual work, and companies investing substantially in AI deployment across verticals. Sam reinforced this point with a striking statistic: enterprise AI adoption has moved from proof of concept to production at a rate approaching 70% in 2025.

The strategic shift: from models to applications

One of the most fascinating strategic discussions centered on where the real value is being created in the AI ecosystem. As Anamitra pointed out, while Afore Capital has not invested in infrastructure or model companies which need to raise billions, they are seeing tremendous opportunity in the application layer.

“The money might be in the applications,” Anamitra explained. “We’re seeing this with coding to start, but finance is coming soon, customer support is happening, and we’re talking about manufacturing, supply chain, healthcare, and other industries.” The key insight is that while foundation models are crucial, the real battle is being fought at the application level where companies have direct access to user data, workflow integrations, and context.

This has created what Anamitra described as a fascinating race: will applications remain sticky and reverse integrate to build their own models? Or will the foundation model companies build better applications and unseat the current leaders? Companies like Cursor are already demonstrating this trend, leveraging their user data and workflows to potentially build their own models, reducing reliance on providers like Anthropic or OpenAI.

Sam reinforced this observation, noting that many startups begin by using the best available models, whether closed or open source. But increasingly, they are building their own models in-house for specific tasks, testing them on lower-priority workflows before scaling up. This hybrid approach allows them to reduce dependency on external providers while maintaining access to state-of-the-art capabilities where needed.

The rise of Agentic AI

Sam shared a prescient insight from her work at AWS. Back at the end of 2023, following ChatGPT’s breakthrough, she wrote an internal document predicting that Agentic AI would be the next major shift. That prediction has proven remarkably accurate.

“We’re seeing startups and enterprises adopt agentic technology across domains,” Sam noted, “whether that’s coding, SRE, or other specialized functions.” This is not just about automating tasks. It is about creating AI systems that can reason, plan, and execute complex workflows with increasing autonomy.

Emerging frontiers: 3D, materials science, and robotics

One of the most intriguing predictions from Sam involves the convergence of AI with physical industries. “I think we’re going to see a whole new round of foundation model builders emerge in the 3D space, materials science, and manufacturing,” she explained. “This will converge with robotics and impact media and entertainment, mining, oil and gas, and other traditional industries.”

This represents a significant expansion beyond text and images into the physical world. The implications for industries like manufacturing, mining, and energy could be transformative.

The infrastructure build-out

Barath brought a crucial perspective on the sheer scale of investment flowing into AI infrastructure. “The amount of capital being poured into this space has been mind-boggling,” he observed. “Not just AWS, but power, data center space, all kinds of infrastructure.”

He also highlighted an important maturation process: enterprises are now adopting AI at scale, and we are seeing the natural tension between enterprise expectations and what new AI companies can actually deliver. This mismatch is being worked through in real-time as both sides learn to work together more effectively.

The changing economics of startups

The panel explored how AI is changing startup fundamentals. Thanks to cloud infrastructure and generous credit programs from providers like AWS, the cost of starting a company continues to decline. As Anamitra noted, while vibe-coding makes it easier for non-technical founders to create demos, you still need technical depth to build production-ready systems.

Interestingly, team sizes have not necessarily changed. Most investments still involve one to three founders, with at least a couple having strong technical capabilities. However, companies can now go further on less capital, often replacing junior engineering roles with AI tools.

The example of Gamma, an Afore portfolio company, illustrates this trend perfectly. With just 30 people, they are generating significant revenue building AI-powered presentation software. They bootstrapped growth with profits after their initial $10 million round from Accel, only raising additional capital from Andreessen Horowitz much later, at a $2 billion valuation, on their own terms.

Copyright and IP: a non-issue for code generation

Given the panel’s expertise, I was curious about the intellectual property implications of AI-generated code. Barath’s response was refreshingly direct: “I think it’s a very overrated issue.”

He explained that if you are not leveraging AI tools to build code, you are already behind. While it is true that AI output is not currently copyrightable under U.S. law, this matters less than many people think. “Software companies aren’t suing each other for copyright infringement,” Barath noted. “What’s important is you get outputs and you keep it secret. You’re not sharing your source code with people. You still have IP there.”

This perspective represents a significant shift in how investors and legal experts view AI-assisted development. The concern that once dominated boardroom discussions has largely evaporated. The law may eventually change to make some AI outputs copyrightable, but for now, the practical reality is that trade secret protection and the competitive advantage of rapid development far outweigh theoretical copyright concerns.

The message to startups is clear: use AI tools aggressively, do not worry about copyright issues, and focus on building products that customers want. As Barath put it, “The cat’s out of the bag on that one.”

Looking ahead: predictions for 2026

As our discussion wound down, I asked each panelist for their predictions for 2026. Their responses revealed both the exciting possibilities and potential challenges ahead.

Barath’s cautionary vision: the dark side of agents

Barath offered perhaps the most provocative prediction: “The first real world harm caused by agents will occur.” He painted a scenario that should give everyone pause. What happens when agents are unleashed on financial markets and cause a flash crash? Or when they autonomously buy airplane tickets and inadvertently purchase overpriced tickets for consumers en masse?

“The more you empower them to use computers, do financial transactions, operate machinery, something has got to give at some point,” Barath warned. His prediction is not about whether agents will become powerful enough to do meaningful work, but about the unintended consequences when they do. It is a sobering reminder that as we rush to deploy autonomous systems, we need to think carefully about safeguards and accountability.

Sam’s vision: the convergence of digital and physical

Sam’s prediction centers on convergence. “I think the convergence of 3D physics, audio, material and applied sciences and robotics is going to be fascinating,” she explained. While she acknowledged the timeline is uncertain, given that one day in AI can feel equivalent to a year, she is confident this convergence will reshape everything from media and entertainment to manufacturing.

This represents a fundamental expansion of AI beyond the digital realm into the physical world. The implications span industries: manufacturers using AI to optimize materials, media companies leveraging 3D models for content creation, and robotics systems becoming more intelligent and autonomous. It is a prediction about AI finally breaking free from screens and entering the tangible world around us.

Anamitra’s dream: the AI concierge

Anamitra took us back in time to make his point about the future. He referenced Magic, a company from over a decade ago that Sequoia invested in, which offered a personal concierge service via chat or phone. You could ask them to pick up flowers, shop for groceries, or deliver food. The company did not survive, but the concept was compelling.

“Something like that is possible now,” Anamitra argued. “Wouldn’t it be cool if you had a personal concierge, without paying too much for it, that would do everything humanly possible?” He envisions an AI assistant in everyone’s pocket that can order food, look up recipes, call a cab, buy tickets, and handle countless other tasks autonomously and reliably.

What struck me most about this prediction was Anamitra’s observation that for most consumers, Gen AI interaction today is “just a little bit better than Google search.” It does not fundamentally change lives yet. But a truly capable personal AI concierge would. “Maybe that’s more important than AGI,” he mused. “Personally I would like to have that more.”

While he acknowledged we are “far from it” today, he believes it is coming next year or the year after. The technology exists; it is now about integration, reliability, and making it work seamlessly in everyday life.

The Human Element

One of the most thought-provoking moments came when an audience member asked about where humans will fit in this AI-powered future. Sam offered an optimistic perspective: AI will allow us to be more intentional and creative with our time, both professionally and personally.

Anamitra noted that while some jobs will indeed disappear, hiring is simultaneously increasing in many sectors. Companies using AI coding tools are actually hiring more engineers, not fewer. This suggests we may be creating new work streams and opportunities even as we automate existing tasks.

Barath brought it back to fundamentals with his perspective as a parent: “What I try to emphasize to my kids is interpersonal skills, social skills, emotional skills. Just like I don’t need to know my multiplication tables anymore, they’re going to need to know fewer facts. But they’ll always need to know how to relate to people, speak well, and be kind.”

Conclusion

As I reflected on our conversation, what struck me most was the genuine optimism tempered with realism. None of our panelists suggested this transformation would be without challenges. Jobs will be disrupted, business models will change, and we will face difficult questions about the future of work.

But the consensus was clear: this is not a bubble. The demand is real, the value being created is tangible, and the opportunities are immense. As I noted at the end of our discussion, when the internet emerged, no one could have predicted jobs like web developer would exist. The same creative destruction and creation is happening now with AI.

Working with YC companies and other startups on the West Coast, I see this transformation daily. Young founders are building genuinely innovative solutions, often with smaller teams and less capital than previous generations required. The tools are more powerful, the infrastructure more accessible, and the potential more expansive than ever before.

The AI revolution is not coming. It is here. And if this panel discussion revealed anything, it is that the most exciting chapters are still being written.

Watch the video:

Arte Merritt is GTM specialist for Amazon Bedrock for Startups. He is the founder of Reconify, an analytics and optimization platform for Generative AI. Previously, he led the Global Conversational AI partner initiative at AWS. He was the founder and CEO of the leading analytics platform for conversational AI, leading the company to 20,000 customers, 90B messages processed, and multiple acquisition offers. He is a frequent author and speaker on Gen AI, Conversational AI, and data insights. Arte is an MIT alum.

We hosted a Built on Bedrock demo night on November 11th at the AWS Builder Loft in San Francisco.

It was a great night to see the latest applications of Agentic AI from Bay Area startups.

The presentations covered agentic sales automation, marketing intelligence, and security.

Each of the startups who presented works with Amazon Bedrock — Amazon’s fully managed generative AI service.

The evening started with a demo of Amazon Bedrock AgentCore, by Jean Malha. AgentCore is a collection of agentic components that makes it easier to build, deploy, and maintain highly capable AI agents securely, at scale.

Learn more about the startups and watch the demo video below.

Presenters

Bedrock Data

Bedrock Data, presented by Bruno Kurtic and Jeremy Linden, is a data security and management platform. The platform discovers enterprise data sets, automatically classifies the data using AI, and builds policies to secure the data and govern access to the data — to ensure the data does not get leaked or misused. Bedrock Data helps enterprises implement responsible AI to make sure data is not exposed via agents.

BlueOcean

BlueOcean, presented by Robbie Whiting, is an agentic AI platform for enterprise marketers. A key feature provides competitive intelligence on competitors — analyzing 2,000 to 3,000 data points collected from public sources weekly, or monthly, by the platform on the company and competitors. This data forms a context engine, that can be queried like a RAG solution. Customers can ask how they compare versus competitors, get recommendations, and generate a complete marketing strategy in minutes that normally would have taken three weeks to three months, if done manually.

HiddenLayer

HiddenLayer, presented by Tom D’Aquino, is a platform that enables customers to test, evaluate, and secure their agentic applications and agents, from a security perspective. The platform features four components: discovery to perform asset inventory on what is running; supply chain analysis to identify threats that have access to model artifacts; runtime protection including guardrails; and security testing, including red-teaming.

Reevo

Reevo, presented by Clement Fang, is an agentic, sales automation platform. The platform automates and augments prospecting, outreach, and pipeline management. The platform can transcribe sales calls and draft emails based on the context of the meeting. Users can also interact with Reevo via Slack and create workflows via natural language.

Sybill

Sybill, presented by Soumyarka Mondal and Mehak Aggarwal, is an agentic-AI, sales assistant. Sybill analyzes all of the conversational data across sales calls, emails, and CRM notes. The platform can determine the stakeholders on a call and create individual emails tailored to each stakeholder’s persona. The before and after results of customers using Sybil showed that sales associates executing in the 75th percentile, in terms of deal velocity and revenue closed, are now executing in the 90th percentile.

Watch the demos

Next Event

Our next event is December 16th at the AWS Builder Loft in SF. Register to attend.

Arte Merritt is GTM specialist for Amazon Bedrock for Startups. He is the founder of Reconify, an analytics and optimization platform for Generative AI. Previously, he led the Global Conversational AI partner initiative at AWS. He was the founder and CEO of the leading analytics platform for conversational AI, leading the company to 20,000 customers, 90B messages processed, and multiple acquisition offers. He is a frequent author and speaker on Gen AI, Conversational AI, and data insights. Arte is an MIT alum.