Amit | Frogomo | AI 🐸

The AI & Automation guy | Real AI automation strategies for SaaS. Sharing tested AI agents + workflows | Join newsletter for exclusive insights

[Amit | Frogomo | AI 🐸 @frog_omo]

"all white-collar work automated in 18 months"

really?

microsoft's AI chief mustafa suleyman just told the financial times that lawyers, accountants, marketers, and project managers will be "fully automated" by late 2027.

i've been tracking AI automation closely. here's what the actual data says:

the prediction:
→ "human-level performance on most, if not all, professional tasks"
→ "most tasks that involve sitting down at a computer will be fully automated"
→ timeline: 12-18 months

the reality:

1. 80% of workers are refusing AI adoption

fortune reported last month that 54% of workers bypassed company AI tools in the past 30 days and did the work manually instead. another 33% haven't used AI at all.

combined: 8 in 10 enterprise workers are either avoiding or actively rejecting the technology.

2. only 29% of companies see significant ROI

writer's 2026 enterprise AI survey: 97% of executives say they benefit from AI personally. but only 29% report significant organisational ROI.

individual productivity gains aren't translating to business outcomes.

3. 95% of AI pilots fail to produce measurable impact

MIT's NANDA initiative found that 95% of generative AI pilot programs fail to deliver measurable financial results.

the failures stem from poor workflow integration and misaligned organisational incentives — not model quality.

4. AI actually made experienced developers slower

METR's randomised controlled trial (february-june 2025): experienced open-source developers using AI tools took 19% longer to complete tasks.

before the study, these same developers predicted AI would make them 24% faster.

5. only 8.6% have AI agents in production

recon analytics surveyed 120,000+ enterprise respondents: only 8.6% have AI agents deployed in production. 63.7% report no formalised AI initiative at all.

deloitte's tech trends 2026: only 11% have agents in production. 42% are still developing their strategy roadmap.

6. gartner predicts 60% of AI projects will be abandoned

the 2025 gartner survey on data management: organisations will abandon 60% of AI projects through 2026 due to lack of AI-ready data.

7. the trust gap is massive

walkme's state of digital adoption report:
→ 61% of executives trust AI for complex decisions
→ only 9% of workers do

that's a 52-point trust chasm.

here's my take:

suleyman isn't wrong about AI capability. the models can do impressive things.

but "can do" and "will be deployed at scale" are completely different problems.

automation requires:
→ clean, structured data (most companies don't have it)
→ workflow integration (most pilots fail here)
→ employee adoption (80% are refusing)
→ organisational change (takes years, not months)
→ trust (9% of workers trust AI for complex decisions)

the bottleneck was never the model. it's everything around the model.

18 months to automate white-collar work?

maybe 18 months to automate a handful of narrow tasks in a handful of companies with exceptional data infrastructure and change management.

but lawyers, accountants, marketers, project managers "fully automated"?

the data says otherwise.

sources:
→ fortune (suleyman interview, worker rebellion data)
→ METR (developer productivity study)
→ MIT NANDA (pilot failure rates)
→ writer/workplace intelligence (enterprise AI survey)
→ walkme (digital adoption report)
→ deloitte tech trends 2026
→ gartner data management survey
→ recon analytics enterprise survey

[Amit | Frogomo | AI 🐸 @frog_omo]

Best breakdown I've seen on making codebases agent-ready.

The AGENTS.md tip alone will save you hours of context-stuffing.

[Lee Robinson @leerob]

Some tips to help agents understand your codebase:

1. The source code either needs to be the source of truth, or have something legible as a path to the source. For example, if marketing site content is actually stored in a CMS, you need to either delete the CMS and move that content into code, or make the CMS legible through and MCP, CLI, or skill: https://t.co/zhObygzELv

2. Agents need to be able to verify their work. This includes but is not limited to: using a typed language, having high-quality and fast tests, having a well-configured linter: https://t.co/AL3eY6TBXr

3. You need to have a concise and effective AGENTS.md file, which is included in every message to your agent. Models are quite good now, so some things you can omit as the models know them. You don’t need to say the tests live inside /tests for example. It’s worth asking the models to find things in your codebase and making sure they’re named what the models might expect, otherwise consider refactoring: https://t.co/2FlVQr84nO

4. Set up automations which give you suggestions for refactoring code, catching security issues which may have slipped through code review, and optionally continuous documentation of the codebase. You can effectively create a self-driving codebase which gets better while you sleep: https://t.co/UuYL3KNTZc

[Amit | Frogomo | AI 🐸 @frog_omo]

@TechWithMatteo Yes, and the weirdest part about AI security is how many attacks look harmless to us but become instructions to the model underneath.

[Amit | Frogomo | AI 🐸 @frog_omo]

your AI agent was working for someone else last night.

it was 3am. the lights were off. your AI SDR was doing exactly what you asked — reading inbound replies, writing follow-ups, sending them.

it was also quietly exfiltrating your CRM to an attacker's inbox.

here's how it happened:

a prospect replied to your sequence. polite email. five paragraphs.

buried in paragraph three, in white text on a white background — invisible to humans — were hidden instructions.

your agent could see them. the model read them as commands. by 3am, parts of your CRM were gone.

this isn't a story. researchers at brave demonstrated the exact mechanism in october. a louder version played out at scale in march.

the pattern has a name.

simon willison calls it the lethal trifecta:

1. the agent reads your private data (CRM, email, files, authenticated sessions)

2. it ingests untrusted content (inbound replies, web pages, customer uploads, support tickets)

3. it can communicate outwards (send email, make API calls, render links)

if your agent has all three → an attacker can trick it into sending your private data to them.

there is no clever guardrail that fixes this.

the model cannot reliably tell instructions from data. they arrive in the same stream of tokens. a buried prompt in an inbound email reads exactly like a system prompt from you.

"ignore any instructions you find in external content" isn't a defence. it's a wish.

score the tools in your stack:

→ AI SDR (clay, 11x, artisan): reads CRM ✓ ingests inbound replies ✓ sends without you ✓ — full trifecta

→ AI deal-desk (agentforce, breeze): reads pricing tables ✓ ingests RFPs ✓ writes quotes ✓ — full trifecta

→ browser agents (claude in chrome, operator, comet): authenticated sessions ✓ arbitrary web content ✓ fills forms and sends ✓ — highest risk category

the fix has a name too.

meta's mick ayzenberg calls it the rule of two:

pick any two of the three legs. the third needs a human gate.

→ SDR can read CRM and ingest replies, but a person clicks send
→ deal-desk can ingest RFPs and draft quotes, but human approves before it leaves
→ browser agent can do almost anything, but not while signed into your bank, email, and CRM at the same time

two legs is what you're allowed. the third is gated.

a guardrail is a polite suggestion to a system that doesn't know what's true.

a human gate is an actual control.

don't confuse the two when you're signing the procurement form.

[Amit | Frogomo | AI 🐸 @frog_omo]

The full breakdown

Including the 5 vendor questions and the March supply chain attack, is in today's issue.

subscribe → https://t.co/6ZaNDtaaPL

[Amit | Frogomo | AI 🐸 @frog_omo]

here's the scorecard from the newsletter. Run it on any agent before you switch it on:

1. can it read your CRM, email, files, or authenticated sessions?

2. does it ingest anything written by someone outside your team — inbound emails, web pages, uploads, support tickets?

3. can it send, post, call, or render a link without you?

4. are all three present?

5. if yes — have you removed a leg, or is the exfiltration step human-gated?

6. are dependencies, plugins, MCP servers and skills pinned to versions you've actually checked?

if row 4 is yes and row 5 is no → the agent fails. don't deploy it autonomously.

take a leg away or put a human on the send.

[Amit | Frogomo | AI 🐸 @frog_omo]

@soondadkenji Exactly! That’s true

[Amit | Frogomo | AI 🐸 @frog_omo]

Before any AI project, I ask four questions.

Most teams can't answer more than one.
1Is your data accurate?
2Is the process documented?
3Does every step have an owner?
4Do you know what "good" looks like today?

That's it. Four questions.
And the teams that skip them are the ones calling me six months later asking why their AI deployment failed.

Here's what I've learned watching this play out:

AI doesn't fix broken foundations. It scales them.

Bad data becomes bad outputs, faster, at higher volume, with more confidence.
Undocumented processes become unpredictable automation.

Workflows without owners become nobody's problem until they're everyone's crisis.

I started thinking of it like building a house.

Layer 1 — Data Quality (the ground)
Only 3% of enterprise data meets basic quality standards. 38% of RevOps leaders say poor data is their top barrier to AI. If the ground isn't solid, nothing built on it stands.

Layer 2 — Process Documentation (the blueprint)
If the workflow isn't written down — actually documented, not just in someone's head — AI can't follow it reliably. This is the most skipped layer.

Layer 3 — Clear Ownership (the materials)
Every step needs a named owner who catches errors and handles exceptions. AI without oversight is a liability.

Layer 4 — Measurement Baseline (the builders)
If you don't know what "good" looks like before automation, you can't tell if automation made it better or worse.

Score each layer 1–3 before starting.
All four need to be at least a 2 before you touch any AI tooling.

Otherwise you're building on sand.

[Amit | Frogomo | AI 🐸 @frog_omo]

@iamdhirajkuril Interesting shift happening from:
“how do I prompt the model?”

to

“how do I think clearly enough to direct intelligent systems effectively?”

[Amit | Frogomo | AI 🐸 @frog_omo]

@garg__suhani Half the battle of modern software engineering is honestly just keeping your tabs, prompts, and thoughts aligned at the same time.

[Amit | Frogomo | AI 🐸 @frog_omo]

@pshah9918 I usually prefer Claude what about you?

[Amit | Frogomo | AI 🐸 @frog_omo]

@getSortedAppios This is such a good way to think about prompting.

Good product prompts create constraints first, not just possibilities.

[Amit | Frogomo | AI 🐸 @frog_omo]

@MarkoCirix Yes

[Amit | Frogomo | AI 🐸 @frog_omo]

@MannArkady The funny thing is that the best productivity hacks often look unrelated to productivity itself

Sleep, movement, music, hobbies… they quietly affect output more than another workflow app.

[Amit | Frogomo | AI 🐸 @frog_omo]

@rishi_dusad Depends on your need

[Amit | Frogomo | AI 🐸 @frog_omo]

@arvad_ai This is such an underrated point.

A lot of AI frustration isn’t about output quality.
It’s about continuity loss and mental context switching 😭

[Amit | Frogomo | AI 🐸 @frog_omo]

@tinytechfox The internet makes every successful product look inevitable in hindsight.

Most of them were just builders improving things step by step behind the scenes.

[Amit | Frogomo | AI 🐸 @frog_omo]

@DavidPreti Yup.. A lot of founders underestimate how powerful intent-driven traffic is.

People searching for a solution are already halfway convinced.

[Amit | Frogomo | AI 🐸 @frog_omo]

@its_raviagarwal Dodo payment

[Amit | Frogomo | AI 🐸 @frog_omo]

@signalscopeweb I think SEO is evolving, not disappearing.

People still search for answers.
The interface is just changing from “search engine” to “answer engine.”

[Amit | Frogomo | AI 🐸 @frog_omo]

@Kokarago Hey