Predictions ledger
What our guests expect to happen
Guests on Chain of Thought often say where they think AI is heading. This page keeps a dated record of those calls: the exact words from the transcript, the timeframe the guest gave, and a status that changes only when a cited source settles it.
- 25 predictions
- 19 guests
- 21 episodes
This is a ledger, not a scorecard. A prediction made on a podcast is a guest thinking out loud, often with caveats that a short quote leaves out, so every entry links to its moment in the transcript. Read the surrounding conversation before leaning on any of them.
Open means an outcome has not yet been recorded here. Other statuses describe what the cited sources show. An entry moves from open only with a public source and the date it was checked. Where a guest gave no timeframe, the entry says so rather than inventing one. Dates next to each quote are the episode's publication date.
Open 21
With a stated timeframe
-
Compute will be hard to come by.
“Actually, and I'm going to tell you this, we're going to have a hard time with, with compute by the end of this year.”
Sterling Chin on EP 50: I Built an AI Coworker That Runs 90% of My Day | Sterling Chin, Postman, Transcript at 30:29 →
-
Energy, more than inference, will be the main limiting factor for AI until new reactors come online.
“I think the really limiting factor is gonna be energy for the next couple years until we can start getting reactors online and really have this up leveling effect.”
Denny Lee on EP 22: AI's Two Extremes – Foundations & The Frontier | Databricks’ Denny Lee, Transcript at 27:11 →
-
Each of an enterprise's software systems will have its own AI agents.
“…but three years from now, you know all of these systems will have their own agents.”
Yash Sheth on EP 16: Information Symmetry: DevRev's Bet on AI-Driven Enterprise Decisions | Manoj Agarwal, Transcript at 20:15 →
-
Organizations will settle on an established pattern for how to approach managing AI agents at work.
“And my guess is that in the next, you know, one to two years max, We're going to kind of establish what this looks like, and there's going to be kind of a pattern of here's how you approach it.”
Tyler Akidau on EP 60: We Built Agents, Nobody Built HR | Tyler Akidau, Redpanda, Transcript at 47:55 →
-
Context engineering will be a role for a few years, then become mechanical and self-learning, as prompt engineering did.
“I honestly think that it's more a role for the next one, two or three years, because a lot of that is going to become very mechanical and self-learning”
Michel Tricot on EP 52: Context Poisoning is Killing Your AI Agents: How to Stop it | Michel Tricot, Airbyte, Transcript at 10:38 →
-
Every piece of software being built will have some form of AI built into it.
“I think in the next three to five years, every piece of software that is built on this planet will have some sort of AI baked into it.”
Atindriyo Sanyal on EP 9: AI in 2025: Agents & The Rise of Evaluation-Driven Development | Galileo's Yash Sheth & Atindriyo Sanyal, Transcript at 18:00 →
-
Critical transactional code, such as banking code, will still need people who understand it.
“But, like, I don't see a path within the next five years to just be like, cool. Nobody's gonna understand my banking code.”
Charity Majors on EP 18: AI Won't Solve Your Toughest Engineering Problems | Honeycomb’s Charity Majors, Transcript at 17:29 →
-
Multimodal AI will erase the language barrier as we know it.
“And it can be three years, it can be five years, but I think language barrier as we know it is going to be erased back to multimodality”
Olga Beregovaya on EP 20: Breaking the Language Barrier: Smartling's AI Translation Pipeline | Olga Beregovaya, Transcript at 38:14 →
-
AI models will keep getting much faster and much cheaper.
“I think we have a very uh strong belief that at least for the foreseeable future, two, three, four years, these models will just continue to get much faster and much cheaper.”
Andy Triedman on EP 79: For AI Agent Fleets, the Model Is Rarely the Bottleneck | Theory Ventures' Andy Triedman, Transcript at 18:11 →
-
AI agents will continue to struggle with tacit, unwritten domain knowledge.
“…which I've seen referred to as tacit information, which is a thing that agents are notably bad at, and they'll continue to be bad at it for the next five years.”
Laurie Voss on EP 74: AI Codes: Product Engineers Decide What to Build | Laurie Voss, Arize, Transcript at 8:15 →
-
Most organizations, particularly in critical infrastructure, will take years to use agentic AI at production scale.
“…it's going to take a little bit for us to get there, you know, maybe a few years, maybe up to a decade … I think it's going to take some time before most organizations can use agentic AI in particular at scale.”
Kris Lovejoy on EP 62: The First Fully Autonomous AI Attack is Coming | Kris Lovejoy, Kyndryl, Transcript at 6:59 →
No end date
-
More people will create software, and the surface area of software will grow a thousandfold.
“I think more people will create software. I think the surface area of software will grow still a thousand X.”
Eiso Kant on EP 23: First Code, Then AGI: Software’s Event Horizon with Poolside Founders Jason Warner & Eiso Kant, Transcript at 45:19 →
-
LLMs will not lead to AGI; a different, non-LLM model architecture will be needed.
“I think LLMs will not bring us to AGI, it might not even be hardware problem.”
Aurimas Griciūnas on EP 38: From Demo to Defensibility: How to Build an AI Business that Lasts | Aurimas Griciūnas, Transcript at 44:19 →
-
About a dozen models will carry most consumer inference volume, while enterprises use tens of thousands of models.
“…but you're not gonna have thousands of models on the consumer side. You will probably have a dozen that'll actually have most of the of the volume of inference. On the enterprise side, I think it's gonna be very different. You might have tens of thousands of models.”
Jeetu Patel on EP 44: The Critical Infrastructure Behind the AI Boom | Cisco CPO Jeetu Patel, Transcript at 18:13 →
-
Many frontier labs will offer interfaces for customers to specialize, fine-tune and post-train their models.
“You're going to see lots of frontier labs providing interfaces to like specialize and fine-tune and post-train, would be my guess, yeah.”
Fergal Reid on EP 49: How Intercom Cut $250K/Month by Ditching GPT for Open Models | Fergal Reid, Transcript at 40:10 →
-
AI token costs are probably at their lowest now and will go up.
“You know, tokens today are probably the cheapest that they're ever going to be and the cost is going to go up.”
Jiaona Zhang on EP 64: Stop Token Maxxing: Find Where AI Actually Pays Off | Jiaona Zhang, Laurel, Transcript at 1:47 →
-
AI agents will use wallets everywhere to complete transactions.
“So on this, I think there is really no doubt, it's really just a matter of time, but agents will use wallets everywhere to finish the transactions.”
Charles Guillemet on EP 65: You Can't Secure an AI Agent with Software | Charles Guillemet, Ledger, Transcript at 15:12 →
-
AI will create more jobs than it removes.
“And I feel like you're going to have more jobs created because of AI in the future, rather than less jobs.”
Jeetu Patel on EP 67: Parenting Your AI Agents for Best Results | Cisco's Jeetu Patel, Transcript at 37:48 →
-
Security and observability will stop being separate domains and fuse into one.
“And as a result, security and observability are going to stop becoming two distinctly different domains. And they actually going to fuse into one”
Jeetu Patel on EP 67: Parenting Your AI Agents for Best Results | Cisco's Jeetu Patel, Transcript at 27:26 →
-
The current AI bubble will pop.
“You can absolutely jump on right now and have a really great time, you know, until the bubble pops, which it will.”
Laurie Voss on EP 74: AI Codes: Product Engineers Decide What to Build | Laurie Voss, Arize, Transcript at 15:59 →
-
AI agents will perform more searches than humans.
“One thing for sure is that agents will be performing more searches than humans for the foreseeable future.”
Saahil Jain on EP 81: Agents Are Overtaking Humans in Search: Why Search Must Be Rebuilt, Transcript at 42:35 →
Happened 1
With a stated timeframe
-
Open-source LLMs will catch up with the best closed-source models of late 2024, such as Anthropic's Claude models.
“…my prediction is by the end of next year, there will be open source LLMs which will catch up with the best closed source models of today.”
Vikram Chatterji on EP 3: GenAI Predictions for 2025 | Databricks & Cohere, Transcript at 1:00 →
Unclear 3
With a stated timeframe
-
At least five to ten AI applications of the kind he says were called wrapper apps the year before will reach $100 million in annual recurring revenue.
“…my prediction is you'll see a lot of these, what used to be called last year, wrapper apps … I think we'll see at least five to ten of those applications reaching a hundred million in revenue in ARR by the end of 2025.”
Vikram Chatterji on EP 3: GenAI Predictions for 2025 | Databricks & Cohere, Transcript at 1:00 →
-
Distributed multi-agent systems, of the kind the A2A protocol is meant to connect, will not be in production yet.
“I think what we will not see, we will not see distributed multi agent systems in production yet, even though everyone is talking about A2A and how it will change the world because companies will start exposing agents as services. Not not in six months.”
Aurimas Griciūnas on EP 38: From Demo to Defensibility: How to Build an AI Business that Lasts | Aurimas Griciūnas, Transcript at 42:07 →
-
Someone will be able to train a reasoning model similar to DeepSeek's on a laptop.
“And at some point, I think in the next, you know, twelve to eighteen months, someone should be able to train a similar model just on their laptop.”
Atindriyo Sanyal on EP 11: How DeepSeek Changed the AI Race Overnight, Transcript at 6:23 →