Dennis Paul

Founder, Cordiant. The Enterprise Customer Memory Company.

[Dennis Paul @Dennispaulm]

@Linkinthedark @BenjaminPDixon @RandPaul Scientist🤡

[Dennis Paul @Dennispaulm]

@canucku2 @helenckh So?

[Dennis Paul @Dennispaulm]

@MartinGTobias The most successful founders in this world have been solo founders with a mission and that outsized hunger to succeed.

Name the world’s most successful founders.

You will realize that you are just riding a standard VC analogy rather than applying first principles thinking.

[Dennis Paul @Dennispaulm]

@MartinGTobias It’s all about positioning.

I just got out of a phase where I finally believe that I’ve positioned an otherwise brilliant product right.

It took me so many iterations, but at the end of it all, the process was well worth it.

[Dennis Paul @Dennispaulm]

@MacaulaysMonkey 🤡

[Dennis Paul @Dennispaulm]

@ThePunnyWorld You were supposed to duck.

Now you’ve egg on your face.

[Dennis Paul @Dennispaulm]

10x productivity gains are definitely there in coding, design and in architecture.

My company gets that, so it’s not something I have heard from someone.

The danger is companies giving away subscriptions to everyone who are designated as developers and asking for results.

The productivity gains do not happen that way.

You need to do this is in a very controlled environment with just 10% of your team. You may not even need others once this 10% gain critical mass.

With respect to applications, the hype with autonomous AI agents is making people try to do stuff what the current state of technology cannot support, both technically and token-cost wise.

As a rule, we do not let AI touch the systems of record. It could be catastrophic to allow them to do that.

We build applications with AI for what it does best in a cognitive layer, and route everything else to good old deterministic code with governance built in.

[Dennis Paul @Dennispaulm]

@anishmoonka With a coding assistant, you don’t need 5000 engineers, you may need 500, may be even lesser.

Putting a new technology on an old landscape is not very clever.

And blaming the technology for that is stupid.

[Dennis Paul @Dennispaulm]

@HedgieMarkets I would say, Microsoft has too many people using Claude Code.

They need far fewer.

Their enterprise hasn’t caught up with the times.

[Dennis Paul @Dennispaulm]

@jjen_abel We’re painfully recovering from one such demo.

[Dennis Paul @Dennispaulm]

Agents have to be built and deployed with a lot of restraint and discipline or we are going to see a lot of disasters happening.

It’s not a tool for everyone unless it’s doing something for them personally and the stakes are not high.

Enterprises should maintain tremendous amount of control when they deploy ai agents, deterministic governance is an integral part of this process.

I would not allow any agent to touch my systems of record, that’s exclusive territory for good old deterministic programs, not ai agents that invariably hallucinate often and a host of other problems.

[BURKOV @burkov]

If you don't understand this, you will not understand why LLM-based agents are irreparably failing for a general-purpose problem solving.

An agent (by the way it was the topic of my PhD 20 years ago) to be useful, must be rational. Being rational means to always prefer an outcome that results in the maximal expected utility to its master/user.

Let’s say an agent has two actions they can execute in an environment: a_1 and a_2.

If the agent can predict that a_1 gives its user an expected utility of 10, and a_2 gives an expected utility of -100, then a rational agent must choose a_1 even if choosing a_2 seems like a better option when explained in words. The numbers 10 and -100 can be obtained by summing the products of all possible outcomes for each action and their likelihoods.

Now here is the problem with LLM-based agents.

The LLM is not optimizing expected utility in the environment. It is optimizing the next token, conditioned on a prompt, a context window, and a training distribution full of examples of what helpful answers are supposed to look like.

Those are not the same objective.

So when we wrap an LLM in a loop and call it an “agent,” we have not created a rational decision-maker. We have created a text generator that can imitate the surface form of deliberation.

It may say things like:

“I should compare the expected outcomes.”
“The best action is probably a_1.”
“I will now execute the optimal plan.”

But the internal mechanism is not selecting actions by maximizing the user’s expected utility. It is generating a continuation that is statistically appropriate given the prompt and prior context.

This distinction matters enormously.

For narrow tasks, the imitation can be good enough. If the environment is constrained, the actions are simple, and the success criteria are close to patterns seen in training, the system can appear agentic.

But for general-purpose problem solving, the gap becomes fatal.

A rational agent needs stable preferences, calibrated beliefs, causal models of the world, the ability to evaluate consequences, and the discipline to choose the action with maximal expected utility even when that action is boring, non-linguistic, or unlike the examples in its training data.

An LLM-based agent has none of that by default.

It has fluency. It has pattern completion. It has a remarkable ability to compress and recombine human text. But fluency is not rationality, and a plausible plan is not an expected-utility calculation.

This is why these systems so often fail in strange, brittle, and irreparable ways when given open-ended responsibility.

They are not failing because the prompts are insufficiently clever.

They are failing because we are asking a simulator of rational agency to be a rational agent.

[Dennis Paul @Dennispaulm]

@allenanalysis Nothing to do with AI.

All to do with getting AI to do things it should not go anywhere near.

[Dennis Paul @Dennispaulm]

You are letting AI agents do things that they shouldn’t be going anywhere near.

AI agents circa 2026 have their role well cut out and that does not include updating a production database, leave alone deleting, which has happened, unfortunately.

Hype is one thing, governance is the key aspect missed here.

I wouldn’t blame the model or anyone else other than you for this.

Early days, still time for course correction. Come out of the hype and make AI do wonders without ever creating situations like this.

[Nikesh Arora @nikesharora]

The false positive rate for LLMs is about 30%, so as of now no model can validate those and to deploy models to assess threats they need to be merged into the inline sensors - they have taken a lifetime to deploy, endpoints,agents (old world), browsers, firewalls on prem and cloud, models can't enforce, nor write enterprise policy which is customer specific.

[Dennis Paul @Dennispaulm]

@gokulr It works the other way too.

If your acv is too low, it means something that can be easily built by an internal team.

Eliminated a vendor. A big win for a mid level or junior employee.

[Dennis Paul @Dennispaulm]

Software has long been GUI-first. We inverted it: conversational-first, with GUI invoked only where deterministic precision is required.

The employee training manual is dead. So is the onboarding period. And so, eventually, is most enterprise software as it exists today.

[Dennis Paul @Dennispaulm]

You need the fewest of people to build world class applications today from scratch or to maintain existing applications.

Anyone thinking otherwise is living in la-la land.

Your founder friend is not under stress, he’s seeing the hand of God working for him.

All other things are noise. Ignore.

[Aravind @aravind]

@jack glad the biggest woke programming called twitter was made into x where we can program ourselves like the way we want now.

[Dennis Paul @Dennispaulm]

In India, a single 400 mg Albendazole tablet typically costs between $0.06 and $0.10 USD (approximately ₹5.60 to ₹9.12 INR) as of February 2026.

Prices for consumer packs (strips of 1 tablet) are tightly regulated by the National Pharmaceutical Pricing Authority (NPPA), which has set a ceiling price of ₹9.12 ($0.10 USD) per tablet.