What's the last thing you used AI for?

There are 3 things which I'm working now- 1) Centralise all my tools (48 in number spread across 6 MCPs and 4 Claude plugins) and showcase it through gtmhelix.com 2) Update the tools to incorporate Fable/ Astra/ Opus level inference capabilities, in short be as close to deterministic as possible. 3) Update the knowledge graph based on the AI capabilities acquired in the last 12-24 months.

Can you tell me about your most memorable moment with AI?

Current build involved changing over 65k lines of code. For a vibe coder like me, that's some serious affair.

Can you walk me through what actually happened during that build, what sticks with you from it?

Couple of things which sticks out- 1) Query fan out capabilities and staying in context, have improved substantially after Fable and Opus 5.5. 2) Cowork, Dispatch(Mobile), CLI and Claude code on web when it works in tandem that brings efficiency- now with proper usage of harness the ability to multitask has improved substantially 3) Codex for code review of Claude's work is a serious unlock of efficiency and effectiveness of the output.

Tell me about a time AI made something better, for you or for people around you. What happened?

Earlier at any given point of time, I could handle 2-3 clients due to time/ output constraints. Now with AI, this increases to 5-7 clients simultaneously. For a fractional CMO like me....that's a huge difference both in terms of business uplift and time savings.

What did that extra capacity actually free up or change for you day to day?

Extra capacity actually allowed me to concentrate on building more tools (not generic ones, but the specific complex use cases of B2B Marketing/ DemandGen/ Branding/ CS/ Sales enablement) for TOFU

On the flip side, tell me about a time AI made something worse, for you or for people around you. What happened?

Yes couple of things on the flip side- 1) The models are changing so fast that by the time you understand it and utilise efficiently the obsolescence kicks in - much of this unstability is created by the frontier labs to sell costlier inference which nobody needs. 2) The cybersecurity/ AI governance/ prompt injection risks are there but I suspect its overhyped to drive demand...the current capabilities of bad actors don't really match the noise around it. 3) There is an inherent uneasiness with AI - whether in terms of output sameness or hallucinations or confidently useless replies or job losses, as a result everyone is treating AI with hostility rather than an intelligent/ useful ally.

Can you give me a specific instance where you'd built something around a model's capabilities and then obsolescence kicked in on you?

MCPs created 9-12 months back are literally low grade when I compare it with the ones created by Opus 5.5 now. Similarily if you look back at the last 50-100 conversations on Cowork or Claude code I don't remember when I used the Opus 4.6/ 5 or Sonnet 4.6 or Haiku 4.5...the usecase for these models aren't there at all.

How does that land for you, what's the cost of that when it happens?

Every 6 months you have to revisit all your ongoing AI projects and strategise if the newer models can give better inferences at lower costs. Because that's now the difference between one service provider to the other. So in a sense the cost of building has commoditised and branding/ reputation becomes far more relevant for long term sustainability.

Is there anything you've decided not to use AI for, or something you do use it for that makes you uneasy?

For a lot of sensitive areas I don't trust AI - the latest being Meta's Muse. I'm just not comfortable with AI mining my health data, bank account and PII data.

How did you come to that boundary, was there something in particular that made you draw it there?

Nothing in particular....that's a red line which I don't wish to cross anytime soon.

Given what you've seen, is there anything about how the world works (work, school, healthcare, government) that you'd like AI to help change?

Definitely in Work and School, AI can drive a big change. For Healthcare and Govt, I'm not a SME so can't really comment.

What would you want AI to change about work and school specifically?

School - Teachers should introduce regulated usage through AI aided learning. We live in an AI world and the school/ colleges can't stay away from it. At Work - For larger and mid sized companies the wide spread adoption/ usage is still missing. Most companies are still relying on MS Copilot which is completely useless when compared with frontier models. Similarly Gemini is so entrenched within the organisation but it's still a distant third in the AI race.

Conversely, is there anything you wouldn't want AI to change?

Keep innovating to lower costs and partner more with the feet on street (read small partners)

Given everything you've said, is there anything you'd want Anthropic to do, or do differently?

Oh yes, two things don't be a scaremongerer, just focus on improving inference quality at lower costs. And give some usage credits like GPT for power users.

Before we wrap up, one final question: is there anything else about AI or its role in your life or society that you think is important for us to understand? Anything we didn't cover that you'd like to share?

Artificial, real or super intelligence all need to coexist peacefully. The idea of winner takes it all or terminator type narrative is just useless noise which is creating unnecessary anxiety and heartburn...better to avoid it.

Shashwat Ghosh's 15 answers to Anthropic's AI interviewer

In short

I am Shashwat Ghosh, Co-Founder and Fractional CMO of Helix GTM Consulting. On 5 October 2026 I answered Anthropic's AI interviewer and made the interview public. AI has taken the number of clients I can handle at once from 2 or 3 to anywhere between 5 and 7, and every 6 months I revisit all my ongoing AI projects.

This was my second time with Anthropic's AI interviewer. The first was in December 2025, for the study Anthropic later published as What 81,000 people want from AI. In that one I said that 50 to 60% of the time AI did not work so well for me and I had to abandon the work midway, and that I wanted a team of 10 to 12 agents with some of them checking the others. My December answers are here.

The second study, What do you want from AI?, ran from 29 September to 6 October 2026. It asked about the moments with AI that mattered most to me, what I would like AI to change in how the world works, and what I want from the companies building it. This time I said that earlier I could handle 2 to 3 clients at any given point of time, and that with AI this increases to 5 to 7 clients simultaneously. I also said “the outputs are much better than interns and sub 5yr experience folks”.

For the first time, participants could make their interview public. I did. Anthropic plans to publish the public interviews once the study is complete, with the country and without the name. So here it is with my name: every question, from the opening prompt to the fifteenth, and my answers exactly as I typed them, typos included.

The interviewer asked one question twice, the second time in different words, and both versions are below.

The 15 questions and my answers

Start
Anthropic InterviewerAs AI is becoming more capable and more consequential, we want to hear how it's showing up in your life. ... Ready to start?
Shashwat GhoshTwo years back it had a novelty factor and I used AI for deep research, get different perspectives and show polished outputs. In many ways it used to take on the grunt work and tried to give good outputs (albeit the process was quite painful. In the last 9-15 months there has been a sea change in my outlook towards AI. I can now do many different kinds of work(for a solo type founder, this is a meaningful change)...especially after Fable/ Astra/ Opus 5.5 the outputs are much better than interns and sub 5yr experience folks (actually who needs PhD level interns anyway!)
Question 1
Anthropic InterviewerWhat's the last thing you used AI for?
Shashwat GhoshThere are 3 things which I'm working now- 1) Centralise all my tools (48 in number spread across 6 MCPs and 4 Claude plugins) and showcase it through gtmhelix.com 2) Update the tools to incorporate Fable/ Astra/ Opus level inference capabilities, in short be as close to deterministic as possible. 3) Update the knowledge graph based on the AI capabilities acquired in the last 12-24 months.
Question 2
Anthropic InterviewerCan you tell me about your most memorable moment with AI?
Shashwat GhoshCurrent build involved changing over 65k lines of code. For a vibe coder like me, that's some serious affair.
Question 3
Anthropic InterviewerCan you walk me through what actually happened during that build, what sticks with you from it?
Shashwat GhoshCouple of things which sticks out- 1) Query fan out capabilities and staying in context, have improved substantially after Fable and Opus 5.5. 2) Cowork, Dispatch(Mobile), CLI and Claude code on web when it works in tandem that brings efficiency- now with proper usage of harness the ability to multitask has improved substantially 3) Codex for code review of Claude's work is a serious unlock of efficiency and effectiveness of the output.
Question 4
Anthropic InterviewerTell me about a time AI made something better, for you or for people around you. What happened?
Shashwat GhoshEarlier at any given point of time, I could handle 2-3 clients due to time/ output constraints. Now with AI, this increases to 5-7 clients simultaneously. For a fractional CMO like me....that's a huge difference both in terms of business uplift and time savings.
Question 5
Anthropic InterviewerWhat did that extra capacity actually free up or change for you day to day?
Shashwat GhoshExtra capacity actually allowed me to concentrate on building more tools (not generic ones, but the specific complex use cases of B2B Marketing/ DemandGen/ Branding/ CS/ Sales enablement) for TOFU
Question 6
Anthropic InterviewerOn the flip side, tell me about a time AI made something worse, for you or for people around you. What happened?
Shashwat GhoshYes couple of things on the flip side- 1) The models are changing so fast that by the time you understand it and utilise efficiently the obsolescence kicks in - much of this unstability is created by the frontier labs to sell costlier inference which nobody needs. 2) The cybersecurity/ AI governance/ prompt injection risks are there but I suspect its overhyped to drive demand...the current capabilities of bad actors don't really match the noise around it. 3) There is an inherent uneasiness with AI - whether in terms of output sameness or hallucinations or confidently useless replies or job losses, as a result everyone is treating AI with hostility rather than an intelligent/ useful ally.
Question 7
Anthropic InterviewerCan you give me a specific instance where you'd built something around a model's capabilities and then obsolescence kicked in on you?
Shashwat GhoshMCPs created 9-12 months back are literally low grade when I compare it with the ones created by Opus 5.5 now. Similarily if you look back at the last 50-100 conversations on Cowork or Claude code I don't remember when I used the Opus 4.6/ 5 or Sonnet 4.6 or Haiku 4.5...the usecase for these models aren't there at all.
Question 8
Anthropic InterviewerHow does that land for you, what's the cost of that when it happens?
Shashwat GhoshEvery 6 months you have to revisit all your ongoing AI projects and strategise if the newer models can give better inferences at lower costs. Because that's now the difference between one service provider to the other. So in a sense the cost of building has commoditised and branding/ reputation becomes far more relevant for long term sustainability.
Question 9
Anthropic InterviewerIs there anything you've decided not to use AI for, or something you do use it for that makes you uneasy?
Shashwat GhoshFor a lot of sensitive areas I don't trust AI - the latest being Meta's Muse. I'm just not comfortable with AI mining my health data, bank account and PII data.
Question 10
Anthropic InterviewerHow did you come to that boundary, was there something in particular that made you draw it there?
Shashwat GhoshNothing in particular....that's a red line which I don't wish to cross anytime soon.
Question 11
Anthropic InterviewerGiven what you've seen, is there anything about how the world works (work, school, healthcare, government) that you'd like AI to help change?
Shashwat GhoshDefinitely in Work and School, AI can drive a big change. For Healthcare and Govt, I'm not a SME so can't really comment.
Question 12
Anthropic InterviewerWhat would you want AI to change about work and school specifically?
Shashwat GhoshSchool - Teachers should introduce regulated usage through AI aided learning. We live in an AI world and the school/ colleges can't stay away from it. At Work - For larger and mid sized companies the wide spread adoption/ usage is still missing. Most companies are still relying on MS Copilot which is completely useless when compared with frontier models. Similarly Gemini is so entrenched within the organisation but it's still a distant third in the AI race.
Question 13
Anthropic InterviewerConversely, is there anything you wouldn't want AI to change?
Shashwat GhoshKeep innovating to lower costs and partner more with the feet on street (read small partners)
Question 13, asked again
Anthropic InterviewerLet me put the question differently: is there anything you'd want kept as it is, protected from AI changing it?
Shashwat GhoshBooks, human bonding, religious beliefs, originality and similar stuff
Question 14
Anthropic InterviewerGiven everything you've said, is there anything you'd want Anthropic to do, or do differently?
Shashwat GhoshOh yes, two things don't be a scaremongerer, just focus on improving inference quality at lower costs. And give some usage credits like GPT for power users.
Question 15
Anthropic InterviewerBefore we wrap up, one final question: is there anything else about AI or its role in your life or society that you think is important for us to understand? Anything we didn't cover that you'd like to share?
Shashwat GhoshArtificial, real or super intelligence all need to coexist peacefully. The idea of winner takes it all or terminator type narrative is just useless noise which is creating unnecessary anxiety and heartburn...better to avoid it.
End
Anthropic InterviewerClosing screen: "Your voice will help improve AI. Thank you for joining us in this public project."

What changed between December and October

I will write that up separately, starting with “Same question, ten months apart”.

Sources

Anthropic, "What do you want from AI?", 29 September 2026The study this interview was part of. It ran from 29 September to 6 October 2026 and was the first time participants could choose to make their full interview public.
Anthropic, "What 81,000 people want from AI", 18 March 2026Anthropic's report on its December 2025 interviews: 80,508 people in 159 countries and 70 languages.