Alex Kehayias

Seeking founders and builders to beta test the Brand Strategy Canvas Claude plugin. Try it 👉 https://lnkd.in/gBgNTp7e Spend ~30 minutes with the agent and walk away with a clear, differentiated positioning statement + key messages to start telling the story of your project. Why does this matter right now? When the tools to build are everywhere, what separates breakout companies is clarity — about who they're for, what they genuinely offer, and why it matters. That clarity can't be generated. It has to be discovered by a founder willing to do the strategic work. This framework guides you through it. I wrote a book on the Brand Strategy Canvas (find it on Amazon!) before starting Orbit, but with the explosion of new projects competing for attention, brand strategy has never been more important — or more overlooked. If you're building something, I'd love your feedback 🙏

20

3 Comments

How can you make sure orchestrating agents across many frameworks doesn't cost you a fortune? Today, Caelum Forder and I look into agent sessions from Coral Protocol.

16

My two cents on the Claude Squeeze Application layer AI companies being built right now will not survive if they incorporate human in the loop as a core part of how work gets done. If your agent requires a human to approve the output, you’re building a better UI for Claude. If your agent helps an accountant do more accounting or a lawyer do more lawyer-ing, you might have a business, but companies won’t pay a lot for it. It’s not that Claude is going to steal your customers. It’s that the delta between what you’re offering and what someone gets from just using Claude directly is getting smaller every month. The complexity you’re solving for, better UX, smarter prompts, cleaner folder structures, is real, but it’s not worth a big check. And buyers are figuring that out. I’m already seeing it. A real estate brokerage looking at AI underwriting tools realized they can dump a rent roll straight into Claude and get 80% of the output for free. They’ll pay a little more for a better interface, but not a lot more. If you’re not doing the job end-to-end, your value is the productivity layer. And Claude is coming for the productivity layer. Hot take: I don’t think Harvey would get funded in 2026. They were funded when they were because the models weren’t advanced enough to reveal how thoroughly they’d eat the application layer. Harvey may survive, they’ve built real brand and enterprise relationships. But with that exact pitch today I think you’d struggle. Here’s what I think survives: Agents that own the outcome. They have to clear every one of these bars: 1. Fully autonomous. No human in the loop. The agent does the job, not assists with it. 2. Accountable for outcomes. Not activity. Results, and the agent is measured on that basis. 3. Mission-critical metrics. The outcomes it owns show up in the Monday morning operating review. The numbers the CEO reports. Going back to Harvey, if they said we’ll just do your legal work for you, end-to-end, accountable to outcomes, that’d be a defensible business in 2026. But notice what it is: outsourced labor with AI as the delivery mechanism. Not a productivity tool. A law firm that doesn’t charge law firm rates. Most AI companies won’t make that leap because it means owning outcomes. It’s hard. That’s what we’re building at Uniti. If you can autonomously run the entire leasing funnel for a property manager and own their rentals-per-month and conversion rate, you’re not a software vendor. Then do the same for all their workflows. You’re outsourced labor. Completely different moat. The mistake is selling the tool when they should be doing the job. If there’s a human in the loop, the customer is paying for the human’s time and your software. As models get better, the human becomes the cost center and your product becomes the overhead. The comparison isn’t your tool vs. Claude. It’s your agent vs. a full-time hire. That’s a different conversation, and a new category.

62

24 Comments

I wrote a long form piece on individual agency and team dynamics in startups. I've been thinking a lot about this, and I think its an important sea change. https://lnkd.in/gsCiPCM6 Preview: I think the abilities that models develop along different vectors - as measured through various benchmarks - interact in non-linear ways that aren't always obvious. For example, if a model excels at coding but struggles with long context windows, it's going to have a very difficult time understanding a large codebase or a problem that's connected to many conceptual areas within your codebase. If it's bad at tool use, it's going to have a difficult time utilizing semantic search to make the intuitive leaps - or what we think of as intuitive leaps - that humans make to go find specific concepts within the codebase. This matters for making sure you're being DRY, not reinventing concepts that are already solved elsewhere, following standards in your own codebase, and so on. Hat tip to Anthropic's research around interpretability [16] - I think it's extremely important for the industry to understand why and how emergent behavior is happening. Think of where we were a year ago compared to now. It's no longer single one-shot chat interface style editing, but rather very extended multi-shot, think-time compute-driven, rule-building, sub-agentic processes. The tooling in software development has gotten so much more advanced than almost any other field, and these tools are building up this massive ability to amplify your own impact. These high-context individuals who may have previously gone down a fairly narrowing track to scale their effectiveness across an organization now have to measure their opportunity cost in a different way. They need to look at evolving new structures around themselves to uninhibit their throughput so the organization can take advantage of how much these individuals can produce in a short amount of time.

17

3 Comments

U.S. energy demand is surging — don’t get left behind. New gas projects and large industrial loads are reshaping power markets at breakneck speed. In this high-stakes landscape, timely and accurate data is essential. Halcyon’s Data Products, powered by a regularly updated authoritative data catalog of 4.5+ million documents from all 50 public utility commissions, ensure you always have the latest energy market intelligence. Download this guide to see how AI can help power sector stakeholders find new growth opportunities first.

6

For the first time in US History, it's now possible to search for land based on exactly what you intend to develop. Introducing SearchAI Intentions, by Prophetic. When we started building Prophetic, one of the things that frustrated us most about land search was the invisible prerequisite: before you could find a site suitable for your townhome, hotel, data center, warehouse, etc. project, you had to already intimately know the market. You had to know the zone codes. You had to know which municipalities allowed what uses. You had to translate your vision into zoning language and then go hunting. If you didn't have that local expertise, you were starting from scratch or waiting for a broker to call you. That has been true for every developer, in every asset class, since the beginning of the industry. We just changed how the industry works - nationwide. SearchAI Intentions now lets you search by exactly what you want to build with 120+ development type categories, any market, any jurisdiction, and results in seconds. A hotel developer searches "Hotel - Full Service". A data center developer searches "Data Center". A senior living operator searches "Senior Living - Active Adult" or "Senior Living - Assisted Living". SearchAI reads the actual zoning codes across every zone and jurisdiction in your search area and returns every viable parcel, ranked by fit. It cites the specific municipal code section behind every result so you can act with confidence. What this means practically: for the first time, hotel developers, energy companies, warehouse and logistics operators, data center site selectors, cell tower companies, senior living operators, medical facility developers, commercial and retail developers, economic development agencies, and major brokerages have a land search tool built for how they actually think about acquisition. They can find sites that meet their exact criteria, in any market, in seconds - instead of literal weeks. If you have ever spent days in county GIS systems trying to find viable sites for a project, I would love to show you what this looks like now. For a deeper look at how it works and how Prophetic unlocks development across all verticals, read the full breakdown here: https://lnkd.in/gPzJbt3w #PropTech #LandAcquisition #RealEstateDevelopment #SearchAI #Prophetic #DataCenters #SeniorLiving #EnergyInfrastructure #IndustrialRealEstate #HospitalityDevelopment #homebuilding #cre #realestate #commercialrealestate #realestatebrokerage https://lnkd.in/gz-zrt2P

82

10 Comments

Kimi K2.5 just dropped. And the timing couldn't be better. Anthropic has started banning Max plan users from running Claude models through Moltbot (formerly Clawdbot). The very tool that made agentic workflows accessible to builders like us. So what do we do? We find alternatives. And Kimi K2.5 is a compelling one. What makes it different: it was trained specifically to decompose complex tasks and orchestrate massive parallel execution. Up to 100 sub-agents. Up to 1,500 coordinated tool calls. 4.5x faster than single-agent setups. This isn't just prompt engineering. They developed something called Parallel-Agent Reinforcement Learning (PARL) to train the model to self-direct agent swarms without predefined roles or hand-crafted workflows. That's agentic in the truest sense: → You give it a goal → You give it tools → It decides how to break down the problem, spawn sub-agents, and coordinate execution On agentic benchmarks, it's currently the strongest: → HLE with tools: 50.2% (GPT 5.2: 45.5%, Claude Opus 4.5: 43.2%) → BrowseComp: 74.9% (GPT 5.2: 57.8%, Claude Opus 4.5: 59.2%) → DeepSearchQA: 77.1% (beating all frontier models) → Seal-0: 57.4% (GPT 5.2: 45.0%, Claude Opus 4.5: 47.7%) Coding is competitive too. SWE-Bench Verified: 76.8%. Not quite Claude's 80.9%, but close. And the cost? About 10x cheaper than Claude Opus 4.5. → Kimi K2.5: $0.60 input / $3 output per million tokens → Claude Opus 4.5: $5 input / $25 output per million tokens It's open weights. 1T parameters, 32B active. You can download it if you have the hardware, or use their API. The best agentic models aren't necessarily the most expensive ones. They're the ones built with agentic architecture as a first-class citizen. And this one is perfect for Clawdbot / Moltbot!

4

I just published my weekly reflections: https://lnkd.in/gExJi8MD Safe as paintball. Infinite focus. Context rot. The betrayal of dumb tools becoming smart. Builders vs coders. Fossilized data. The mask-off moment for aggregators. Reclaiming your data. Cooperation as transcendence. The Bright Ages. LARPing productivity. The candid aim test. Social alchemy. 

19

Growth Engineers optimized experiments. Activation Engineers optimize systems. The shift matters because AI does not want experiments. It wants feedback loops. Activation Engineers design invite flows, onboarding sequences, and collaboration surfaces that learn over time. The product gets smarter as usage grows. Vortex exists because no one company wants to build this stack in house.

6

I'll post more about this later, but for startups looking for an ai boost to get your team aligned and moving, check out Travis Kahn's Growth Blueprint: https://lnkd.in/e4bcd_dG. Founders and teams pair with an agent to get real foundational work done - Business Model Canvas, Customer Profiles, Messaging Frameworks, and Go-to-Market in a centralized environment that shows how everyone's effort connects (and hopefully amplifies). This is very similar to the strategy work that I did in previous lives at BrandNewMatter and Algoworks, and it really shows me how fast ai is moving. The important thing to understand is that frameworks and platforms like this and r4 Technologies are powerful because the technology works with the PEOPLE you have in the loop. True force multipliers only occur when tech, data, and staff are aligned and empowered.

10

1 Comment

Aways fun talking AI and Bio with Andrew Dunn from Endpoints News! We cover a lot of ground, but there's one persistent theme: 𝗔𝘀 𝗔𝗜 𝘁𝗶𝗺𝗲𝗹𝗶𝗻𝗲𝘀 𝗮𝗰𝗰𝗲𝗹𝗲𝗿𝗮𝘁𝗲, 𝘁𝗵𝗲 𝗴𝗮𝗽 𝗯𝗲𝘁𝘄𝗲𝗲𝗻 𝗧𝗘𝗖𝗛𝗡𝗢𝗟𝗢𝗚𝗬 𝗮𝗻𝗱 𝗽𝗮𝘁𝗶𝗲𝗻𝘁 𝗜𝗠𝗣𝗔𝗖𝗧 𝘀𝗲𝗲𝗺𝘀 𝗼𝗻𝗹𝘆 𝘀𝗲𝗲𝗺𝘀 𝘁𝗼 𝗴𝗿𝗼𝘄. Some thoughts on what that means and what we can do about it…   𝗧𝗿𝗮𝗻𝘀𝗹𝗮𝘁𝗶𝗼𝗻𝗮𝗹 𝗔𝗜 - There’s a mismatch between where AI is working well (early discovery, lead optimization) + the big drivers of risk/cost in drug discovery (target selection, toxicity, clinical, commercial) [1]. But that's changing with AI progress in target discovery [2], toxicology [3] and new AI tools for clinical development [4] and commercial pharma [5] 𝗦𝗵𝗶𝗳𝘁 𝗟𝗲𝗳𝘁 - As AI capabilities shift right, there's an opportunity to incorporate them earlier in the process as a filter for early discovery. We're already seeing protein models bake developability into sequence generation [1] and synthesizability-guided chemical design [2] - but the real unlock will come from stacking properties and integrating them more deeply into the models [3] 𝗧𝗵𝗲𝗿𝗮𝗽𝗲𝘂𝘁𝗶𝗰 𝗔𝗹𝗶𝗴𝗻𝗺𝗲𝗻𝘁 - AI models trained on public literature over-index on academic targets and data from non-human model systems, hurting their performance on real-world discovery tasks. Aligning AI models to human biology and therapeutically relevant problems is a major challenge, but we're seeing progress across the stack -- from few-shot learning [1], in situ human data generation [2], therapeutic proxies [3] and even whole organ screening [4] 𝗖𝗹𝗶𝗻𝗶𝗰𝗮𝗹 𝗦𝘁𝗿𝗮𝘁𝗲𝗴𝘆 - Most promising drugs never see the clinic and many of those that do drop out for external/ macro reasons, creating a buyers market for in licensing. As AI gets better at predicting clinical risk [1], sizing commercial opportunities [2] and accelerating trials [3] we should see an rise of AI-native biotechs [4] rescuing, repositioning and purposing promising late-stage assets where they have differentiated insight and strategy. 𝗠𝗶𝘀𝗮𝗹𝗶𝗴𝗻𝗲𝗱 𝗘𝘅𝗽𝗲𝗰𝘁𝗮𝘁𝗶𝗼𝗻𝘀 - Lastly, we need to reposition AI as an enabling technology, not just a productivity tool. The original sin of TechBio was the promise of cheaper and faster drugs it would never be able to deliver. Instead we should focus on what new capabilities AI is enabling — expanding druggable targets [1], discovery novel biology [2], enabling new therapeutic strategies [3]. Even if this does nothing to improve the cost or speed of a drug approval (which I think it will) it will give us better drugs for more diseases and more patients, which in my book is even ore important! Curious what others are seeing + doing to close the translational gap? https://lnkd.in/daYXihS6

145

14 Comments

Decision modeling is key. Intent + decision traces created by agents need to be modeled to create self learning or ever improving decision quality ( measured via outcomes), while the criteria is non stop shifting. Process modeling - aka crud on system of records will also get eroded and merged with decisions models being stored. This is where Agebt plane and human plane converge.

6

Investors (esp deep tech investors) generally think about things/bet on "layers": the application, the infrastructure, the model. I think that's becoming the wrong way to look at it? The real opportunity is in the feedback loop *between* the layers. A new social app creates demand for new hardware, which in turn demands a new kind of intelligence - I'm calling it "The Agency Loop". My latest newsletter explores this thesis through three signals: the "weaponized transgression" of Cluey, the AI-driven materials science in Nature, and the architectural critiques of Kenneth Stanley. Bet the loop, not the layer :) You can read the full analysis here: https://lnkd.in/ggdVGEda #vibecap #vibecapital #vc #deeptech #AI #cluely

4

1 Comment

Ontologies remain one of the most effective ways to ground AI memory, keeping extracted knowledge anchored to existing structures and adaptable through feedback. As memory systems evolve, their role remains essential, and a core focus at Cognee.

44

4 Comments

i don't think most people have priced in the fact that codegen is now trivial. the bottlenecks are shifting upstream -- why build, what to build -- and, importantly for us, downstream -- whether what's built holds up in the real world. no software will ever be perfect. bugs happen in production. new corner cases uncovered. reality wins. so the differentiator going forward is designing systems that expect failure, then learn from it. when something breaks, it should be detected, root-caused, and fixed -- fast. and the system should remember the issue so that it's more resilient in the future. anti-fragility will be the new baseline.

9

1 Comment

The hottest role in AI startups right now isn’t Forward Deployed Engineers. It isn't GTM Engineers. It’s Deployment Strategists. Decagon calls it an “Agent Product Manager.” Harvey calls it a “Solutions Architect.” Palantir Technologies has had versions of this role for years. And the salaries are climbing fast: - Decagon: $200k–$285k - Palantir Technologies: $120k–$200k - Figma: $150k–$260k - Ramp: $100k–$180k - Harvey: $190k–$260k So who are these people? They are usually pseudo-technical -- CS or engineering majors, or folks with technical work experience. Many come from 2 years in consulting, IB, or PE, then jump into startups to get their hands dirty. They are young, hungry, polished, and comfortable being in front of customers. What do they actually do? They make sure enterprise AI deployments succeed. A $100k+ deal does not survive on a nice pitch or a self-serve onboarding flow. It survives if the customer sees value in the pilot. That means: - Embedding directly with the customer - Designing prompt logic for specific workflows - Working with engineering to align integrations and data flow - Helping exec teams define their AI roadmap - Running feedback loops into product and GTM Why does this role matter so much? Because enterprise AI is messy. Integrations, data transfer, and adoption make or break a deal. Most buyers are using AI for the first time, and each has unique workflows. Deployment Strategists bridge that gap. They own the outcome. They are accountable for making pilots successful, which often means millions in revenue down the line. At Conversion, Sam Bochner has been leading this work for us. We are now thinking about scaling it into a full team. Because a few successful pilots can fund an entire department, and the cost of failed deployments is too high to ignore. Is this just a rebrand of customer success? Not really. Success is about answering tickets and renewals. Deployment Strategy is about going deep with a few enterprise accounts, extracting maximum value, and ensuring the pilot closes into a multi-year contract. Call it Agent PM, Solutions Architect, or Deployment Strategist. Whatever the title, this is becoming one of the most important roles in AI SaaS.

655

91 Comments

Emergent's Testing Agent handles quality control for you, running tests and fixing issues automatically before you even see them. Watch creator Chris Ashby see the difference this makes when building a full-stack app, "this saves me going through and figuring out these issues myself." Our multi-agent system, emulating how a real engineering team works, is how we ensure that apps made on Emergent are fast to build and robust for deployment. #vibecoding #aiagents #aicoding | Source: Build Great Products, YouTube

52

Everyone's interested in understanding why the spec is so important. Watch this talk from Dex Horthy (YC alum, founder of Human Layer) last month to help understand. Dex describes shipping 20,000-line PRs of Go code every few days. Not CRUD apps; these are real production systems with race conditions and shutdown orders. How? By flipping his workflow to spec-first development. He stopped reviewing raw code altogether. Instead, he reviews specs, tests, and plans, and lets agents handle the rest. Some of the key insights from Dex’s process: - A bad line of research can cascade into thousands of bad lines of code. So, the spec being right is essential (read: acceptance criteria) - The best way to scale agents in messy, brownfield codebases is through frequent intentional compaction of context into a spec. The result? They fixed a 300,000-line Rust codebase live on a podcast. They shipped 35,000 lines of production code in 7 hours with a CEO pair-programming. This is what you can do with a spec for your codebase! Are you adopting spec-driven development yet? Watch the full talk here: https://lnkd.in/gTVQuZtw

15