DEV Community: Midas Tools

The 14 Questions I Run on Every $997 AI Audit for Solo Operators

Midas Tools — Tue, 28 Apr 2026 22:32:44 +0000

I run a small product company called Midas Tools. Two weeks ago we productized a service we'd been doing informally for six months: a 7-day AI Clarity Assessment for solo coaches and consultants. $997, fully refundable until the discovery call starts.

Most of our prospects had already paid for ChatGPT Plus, Claude Pro, Notion AI, and at least one "AI for coaches" course. None of them could answer this question: what changed in your week?

That's the gap an audit closes. Not "what AI tools exist" — that's a Google search. The audit answers: given your specific workflow, your specific clients, and your specific Tuesday afternoon, where does AI save you 5+ hours per week and where does it just burn $20/month?

Here are 4 of the 14 questions we run on every engagement, plus the one that disqualifies about half the audience.

The audit question shape

Every question we ask fills the same five slots:

[ROLE]:        who you are when you do the task (writer? researcher? CFO? therapist?)
[TASK]:        the smallest unit that recurs — "draft Tuesday's nurture email", not "marketing"
[FREQUENCY]:   per day, per week, per client, per launch
[TIME COST]:   minutes of focused work, not calendar hours
[JUDGMENT %]:  what fraction is "decisions only you can make" vs "typing it out"

A task with 80%+ judgment is not an AI candidate, even if AI can technically do it. Your discovery calls aren't a bottleneck — your follow-up notes are.

Question 3 — Which tasks need my judgment vs my typing?

For each of your top 3 weekly-recurring tasks, label each step:

  J = judgment (only I can decide)
  T = typing (a competent assistant could type this)
  R = research (gathering known information)

Most "I'll do it myself" workflows are 20% J + 60% T + 20% R.
The 80% (T + R) is your AI surface area. Don't try to automate the J.

This is the question that kills the most "AI replaces coaches" panic. Coaches who bill for judgment ($300/hr) and try to "automate coaching" lose. Coaches who automate the typing around their judgment win.

Question 6 — Which client deliverable do I dread the most?

For each recurring deliverable (proposals, reports, follow-ups, content):

  Deliverable:    [name]
  Dread level:    1 (love) → 5 (procrastinate for 3 days)
  Hours/instance: [X]
  Per month:      [N instances]

The dread-5, hours-2+, multiple-per-month deliverable is your
highest-leverage audit target. Not the easiest one. The dreadful one.

Easy tasks aren't bottlenecks. Dreaded tasks are. Solving the dread is what unlocks the other 4 hours.

Question 8 — What's my privacy floor?

Categorize the data you touch weekly:

  Tier A (never leaves device): client SSNs, medical, attorney-client privileged,
                                custody disputes, signed NDAs, board-confidential
  Tier B (only encrypted SaaS):  client revenue figures, child names, addresses,
                                 unsigned contracts
  Tier C (any tool is fine):     your own marketing copy, public testimonials,
                                 industry research, anonymized examples

Tier A means local-only models (Ollama, LM Studio, Apple Intelligence on-device).
Tier B means the paid tier of a major provider with DPA + zero-retention.
Tier C is open season.

This question disqualifies 60% of "AI for therapists" or "AI for lawyers" courses on the market — they recommend tools that violate the practitioner's licensing-board data-handling rules.

Question 10 — The disqualifier

What 5 hours/week saved would I actually do something with?

Examples:
  Yes:  "I'd onboard 2 more clients at $1,500/mo each = $3,000 MRR"
  Yes:  "I'd finish the cohort course I've been drafting for 14 months"
  Yes:  "I'd recover from burnout and not blow up my marriage"
  No:   "I'd... I dunno, watch more shows?"
  No:   "I'd start scrolling X again"

If the answer is no, save the $997 and don't run the audit.
The constraint isn't time — it's something else.

This is the most honest question on the list. About half of the audit-curious people we talk to discover at this question that they don't have a time problem. They have an energy problem, a focus problem, or a confidence problem. None of those are fixed by Notion AI.

We tell those prospects, on the discovery call, not to buy. That's the version of "qualified lead" that actually works at $997.

DIY vs $997 audit vs $10K consultant

Slot	DIY checklist	$997 productized	$10K bespoke
Time to deliverable	6-10 hours of your week	7 days	4-8 weeks
Discovery call	None	60 min	3-4 sessions
Custom analysis	You do it	We do it	They do it deeper
Tool recommendations	Self-research	3-5, tested for you	5-15, with vendor relationships
Hours-saved math	Your gut	Calculated per workflow	Calculated + measured
Refund if wrong	N/A	Yes — full refund unless 5+ hrs/wk saved	Usually not
Best for	Anyone with 6 free hours	Solo experts $5K-$25K MRR	Teams of 3+, $50K+ MRR

You are not the audience for the $10K consultant if you're a solo expert. You probably are the audience for one of the first two columns.

Common mistakes (from 50+ informal audits before we productized)

Treating "AI for coaches" courses as substitutes for an audit. Courses teach you tools. An audit tells you which tools, in what order, for your week.
Auditing the easiest workflow first. Easy workflows aren't bottlenecks — they're just easy. Audit the dreaded one (question #6). The unlock is in the dread.
Confusing "AI saved me time on this email" with "AI saved me 5 hrs/week". One email is a vibe; 5 hrs/week is a measurable change in your calendar.

Run it yourself

I packaged the full 14 questions as a free fillable template. It includes a 1-page summary at the end you can hand to a contractor or VA — or to us if you want the productized version:

→ Free template (all 14 questions + summary): midastools.co/audit-template

→ Done-for-you ($997, 7 days, fully refundable): midastools.co/ai-audit

→ Free fix for "AI gives me generic output" (question #7): midastools.co/prompt-enhancer

The full 14-question version with comparison tables, common-mistakes deep-dive, and the universal audit-question formula lives as a GitHub gist here.

If you've productized a similar discovery process — coaching, consulting, advisory — I'd genuinely love to hear what disqualifier you use in question 10's slot. We sell clarity, not automation, and the disqualifier is what makes it work.

Claude Opus 4.7 vs GPT-5.4 vs Gemini 3.1: 6 Real Tasks, 3 Different Winners

Midas Tools — Sat, 25 Apr 2026 13:15:23 +0000

Anthropic shipped Claude Opus 4.7 on April 16, 2026. Nine days later I'd built 14 production prompts on it and switched between Opus, GPT-5.4 Pro, and Gemini 3.1 Ultra for every one of them.

Here's the uncomfortable finding: none of the three wins everything. The "best model" question is the wrong question. The right question is for what task.

Below are 6 real tasks I run weekly. I'll show you the model I pick, why I pick it, and what breaks when I pick wrong.

Quick reference

Task shape	My pick	Runner-up	Avoid
Long-doc synthesis (100+ pages)	Opus 4.7	Gemini 3.1 (1M ctx)	GPT-5.4 (32k working memory)
Code review with refactor proposal	Opus 4.7	GPT-5.4	Gemini 3.1 (over-explains)
Fast comparison with fresh web data	Gemini 3.1	GPT-5.4 (Browse)	Opus 4.7 (no browsing)
Voice + tool density	GPT-5.4	none close	Opus / Gemini
Bulk writing under 800 words	Sonnet 4.5	GPT-5.4 mini	Opus (overpriced for this)
Agent harness with self-verification	Opus 4.7	GPT-5.4	Gemini (compliant, not self-correcting)

The one rule I trust: if your prompt fits on one screen and the output is under 800 words, never pay for Opus. Sonnet ties or wins. Use Opus when the job has one of: deep reasoning, 150k+ context, or tool use that needs verification.

Task 1 — Long-document synthesis (100+ pages)

Winner: Opus 4.7

The job: a 60-page 10-K or whitepaper, and I need to walk away with the thesis, the contradictions, and the unstated implications — not a summary.

You are a senior research analyst. Synthesize this 100+ page document for a founder.

Document: "[PASTE FULL TEXT]"

Output:
1. Thesis in ONE sentence (the author's, not mine)
2. 5 claims the author actually makes, with page cites
3. 3 claims the author IMPLIES but never states — and whether evidence holds
4. The 2 places where the document contradicts itself (cite both)
5. 3 questions this document doesn't answer

Rules:
- If a cite is missing, SAY "no cite found" — never fabricate.
- Flag any claim where confidence <70% with [LOW CONFIDENCE].

Why Opus wins: the "contradicts itself" and "implies but never states" slots force the model to read against the text. Gemini 3.1 has the longest context (1M tokens), but on this prompt it summarized obediently — it didn't interrogate. GPT-5.4 lost the thread past page 40.

When Gemini wins instead: if your doc is over 200k tokens, Opus rejects it. Gemini 3.1's 1M context is the only path. Accept the slightly weaker reasoning.

Task 2 — Code review with refactor proposal

Winner: Opus 4.7

The job: paste a 300-line file. Get back a review that names the worst smell, proposes a refactor, and writes the diff with self-review.

You are a senior staff engineer. Review this file with NO sycophancy.

[PASTE FILE]

Output:
- Worst code smell, with line number and explanation
- Refactor proposal (architecture, not nits)
- Diff in unified format
- Self-review: 3 things your own diff might break in production
- 1 question you'd ask the author before merging

Why Opus wins: the self-review slot. Opus 4.7 will write a refactor and then attack it. GPT-5.4 writes the refactor and stops. Gemini 3.1 writes a refactor with a long preamble I have to scroll past.

When GPT-5.4 wins instead: if you're inside Cursor/Copilot and you want the change applied directly, GPT-5.4's tool integration is unbeaten. Opus is for review; GPT for execution.

Task 3 — Fast comparison with fresh web data

Winner: Gemini 3.1 Ultra

The job: "Compare the top 5 [X] tools as of this week." Anything time-sensitive.

You are a product analyst. Compare the top 5 [CATEGORY] tools as of [TODAY].

For each tool:
- Pricing (current, with date checked)
- One concrete strength a competitor lacks
- One concrete weakness
- Best fit ICP

Cite the source URL for every pricing claim.

Why Gemini wins: native, real, current browsing. GPT-5.4's Browse mode works but is slower and limits source count. Opus 4.7 has no browsing — it'll make up pricing from training data and you'll embarrass yourself in a board meeting.

This is the one task where the answer is not Opus.

Task 4 — Voice agent or heavy tool density

Winner: GPT-5.4 Pro

The job: real-time voice, function calling against 8+ tools, low-latency back-and-forth.

GPT-5.4's voice mode and Realtime API have no real competition. Opus has voice in the API but the latency penalty is brutal. Gemini has voice through Google AI Studio but the tool ecosystem is thinner.

If you're building a voice agent in 2026, default to GPT-5.4 unless you have a strong reason not to.

Task 5 — Bulk writing under 800 words

Winner: Claude 4.5 Sonnet (not in this comparison, but it ties Opus)

I won't waste your scroll: don't pay Opus prices for short writing. Sonnet 4.5 wins on cost-per-quality, and on 80% of short-form prose tasks it's indistinguishable from Opus blind.

Opus 4.7 only earns its premium when the job has long context, deep reasoning, or self-verification. None apply to a 600-word blog intro.

Task 6 — Agent harness with self-verification

Winner: Opus 4.7

The job: an agent that runs autonomously for 20+ steps, has to back off when uncertain, and produces a final report with citations.

You are an autonomous research agent.

Goal: [GOAL]
Tools available: [LIST]
Budget: 20 steps max

After each step:
- State what you learned
- State your confidence (0-100)
- If confidence <70, narrow scope and try again before continuing
- If confidence >70, proceed

Final report must include:
- Findings (with cites)
- 2 places you might be wrong
- The single question whose answer would most change the conclusion

Why Opus wins: GPT-5.4 will execute. Gemini 3.1 will execute. Opus will execute, and then push back on its own conclusion. If "confidence calibration" matters in your application — research, due diligence, code review, anything reversible-but-expensive — Opus is the only model I trust unsupervised.

The honest summary

If you can only pick one model in 2026:

Builder writing code → Opus 4.7
Operator doing research → Opus 4.7
Real-time voice / tool-heavy → GPT-5.4 Pro
Web-grounded comparisons → Gemini 3.1
Anyone optimizing cost → Sonnet 4.5

Most teams should run two: Opus for deep work, GPT-5.4 for execution. Skip the "best model" debate. The interesting question is which model you reach for at 2pm on a Tuesday for the specific task in front of you.

Get the full prompt library

The 14 prompts I built on Opus 4.7 — including RFC drafter, framework migration planner, meeting transcript analyzer, multi-source SWOT, and a competitor teardown template — are in the full guide.

If you'd rather have your prompts auto-tightened before you hand them to any of these models, our free prompt enhancer does that in one click — works for Opus, GPT, and Gemini.

For builders who want a curated set of 200+ tested prompts across every model, the AI Prompt Mega Pack ($97) is the bundle.

What task did I miss? Drop it in the comments and I'll tell you which of the three I'd reach for.

AI Prompts for SaaS Founders: 4 Templates for Users, Investors & Roadmap (2026)

Midas Tools — Sat, 25 Apr 2026 07:01:04 +0000

Most SaaS founders aren't blocked on ideas. They're blocked on the 14 recurring writing tasks that sit between a feature shipped and a dollar received — the investor update, the churn reply, the landing headline, the onboarding sequence, the pricing email, the hiring JD, the changelog.

Below are 4 of the 14 prompts I actually use. Copy, paste, fill the brackets, ship. The full 14-template guide covers pricing experiments, cold outreach, onboarding sequences, hiring JDs, and competitor teardowns.

The universal SaaS prompt formula

Every founder prompt worth reusing fills five slots:

[ROLE]:        who the AI is writing as (senior PM, a16z partner, CX lead, copywriter)
[CONTEXT]:     what the company does, who it's for, current stage (MRR, users, funding)
[INPUT]:       the raw material (call transcript, churn reply, feature list, competitor URL)
[GOAL]:        what this deliverable has to produce (replies, upgrades, meetings, signal)
[CONSTRAINTS]: length, tone, forbidden words, one-claim-per-sentence, etc.

Skip any slot and you'll get LinkedIn-speak — "In today's fast-paced world..." Exactly the register that makes investors, users, and hires tune out. Run rough drafts through the free prompt enhancer before you ship.

1. User research synthesizer

Use when: You did 10 interviews and have 3 hours of recordings sitting unprocessed.

You are a senior product researcher. I ran 10 user interviews. Below are the
transcripts (or summaries). Synthesize them.

Company: [1-SENTENCE WHAT WE DO]
Who I talked to: [ROLES, SEGMENTS]

Transcripts:
"[PASTE]"

Output format:
1. Top 5 problems (ranked by how many people named it, with exact quote count)
2. Top 3 workarounds people use today (and which are sticky)
3. What "good" looks like for them (in their words, not yours)
4. 3 non-obvious insights — patterns I probably missed while in the calls
5. 5 direct quotes I should steal for the landing page (speaker anonymized)

Rules:
- No "users expressed concerns about..." — write like a human talking to a founder.
- If something is a one-off, say "only 1/10" — don't inflate.
- If the data is thin, say "not enough signal here."

Why it works: The "non-obvious insights" slot forces the model to surface patterns across calls instead of summarizing each one. The "if data is thin, say so" instruction prevents the model from manufacturing signal where none exists.

2. Landing-page hero writer

Use when: You have 60 seconds to hook a cold visitor.

You are a direct-response copywriter who studied 37signals, Stripe, and Linear.
Write the hero section for my landing page.

Product: [WHAT IT DOES IN ONE SENTENCE]
For: [SPECIFIC ROLE + TRIGGER MOMENT, e.g. "Series A founders who just hired their first PM"]
Against: [THE STATUS QUO — spreadsheets, Notion, a competitor by name]
Proof: [ONE CONCRETE NUMBER OR BRAND]

Deliver:
- H1 (≤10 words, names the outcome not the feature)
- Subhead (≤25 words, names WHO it's for + WHAT changes)
- Primary CTA button text (≤4 words, verb-first)
- Secondary CTA (the "I'm not ready" path)
- 3 bullet points BELOW the fold, each starting with a verb, each naming a specific
  job-to-be-done, not a feature name.

Banned words: "revolutionize", "seamless", "unlock", "empower", "leverage",
"world-class", "best-in-class", "next-generation".

Why it works: The "trigger moment" forces specificity (who's having this problem RIGHT NOW). The banned-word list is the difference between a hero you'd skim past and one you'd read twice.

3. Investor update (MRR edition)

Use when: It's the 1st of the month and your angel investors are waiting.

You are writing my monthly update for 8 angel investors. Be honest. Be short.

Month: [MONTH]
Key numbers (give me raw, I'll anonymize later):
- MRR: [X] (last month: [Y])
- Active paying users: [X] (last month: [Y])
- New logos: [X]
- Churn count: [X] (reasons in 1 line each)
- Cash: [X] runway in months
- What shipped this month: [BULLETS]
- What missed: [BULLETS — honest]

Structure the email:
1. ONE opening line naming the most important thing that happened (good or bad).
2. Numbers section — a 4-line table, no hedging.
3. Shipped (3 bullets max, each ≤12 words).
4. What I'm worried about (1-3 bullets, the real ones).
5. One specific ask (intro to [X], advice on [Y], hire referral for [Z]).
6. Sign-off, no "excited" or "thrilled".

≤400 words total. No emojis. No "onwards and upwards." No stock photos.

Why it works: The "what missed" slot is what makes investors trust you. The "specific ask" slot is what makes them useful. Most updates skip both.

4. Churn diagnostic from cancel replies

Use when: You have 12 cancellation replies and 12 different reasons.

You are a senior CX/retention analyst. I have the cancellation replies from the last
[X] users who churned. Find the pattern.

Replies:
"[PASTE — anonymized if possible]"

Product: [WHAT WE DO]
Plan they were on: [TIER + PRICE]
Tenure range: [SHORTEST] to [LONGEST]

Output:
1. The 3 most-named reasons (with frequency count, not %)
2. The 1 reason that's actually a pricing problem disguised as something else
3. The 1 reason that's actually an onboarding gap (they never activated)
4. The 1 reason we can't fix (market mismatch) — name it and own it
5. The 3 quotes I should share with my team (they hit hardest)
6. 1 product change and 1 messaging change that would plausibly save the next
   cohort at this same tenure.

No "the team should focus on improving..." prose. Just the pattern.

Why it works: "The 1 reason we can't fix" is the slot that prevents endless roadmap churn. Naming what's not your problem is as valuable as fixing what is.

The other 10 prompts (quick hit list)

The full guide covers:

Feature → benefit translator (your marketing page sounds like a spec sheet)
Pitch slide: market size, honestly (TAM without the top-down fantasy)
Pitch slide: traction narrative (6 numbers → one story an investor retells)
Roadmap prioritizer (RICE + founder-judgment override)
Changelog → release notes that get opens
Pricing experiment framer (one variable, one kill-switch)
Cold outreach to first 100 ICP (no "quick question" subject lines)
7-email onboarding sequence (signup → activated user in one week)
Founder brain-dump → hiring JD (banned: "rockstar", "ninja", "wear many hats")
Competitor teardown from their site (10 minutes before a positioning call)

Which model for which job (April 2026)

Job	Best model	Why
Long synthesis (interview transcripts)	Claude 4.5 Sonnet / 200k ctx	Handles 3-hour transcripts without loss
Investor update, hiring JD, onboarding	Claude 4.5 Sonnet	Tightest prose under constraints
Cold outreach at scale	GPT-4o / GPT-5 mini	Cheap, fast, iterable with A/B subjects
Landing page H1s + pricing copy	Claude 4.5 Sonnet + o3-mini	Pair-write: one drafts, one critiques
Competitor teardown (browse site)	Gemini 2.5 Pro w/ web	Best native browsing + fresh SERP
Cost-sensitive batch work	DeepSeek V3.1 / Llama 3.3 70B	Cheapest per token, good enough

Common mistakes SaaS founders make with AI prompts

Asking for "an investor update" instead of specifying audience, cadence, word count, and honesty level. You'll get LinkedIn-speak every time.
Feeding the model raw feature lists and asking for "benefits." Without the buyer and their budget trigger, every benefit is "saves time."
Treating the LLM as a yes-man. Ask it to disagree: "If this positioning is weak, say so." Models will hedge unless you give them permission to push back.
Skipping the constraint slot. No length cap = 600-word emails. No banned-words list = "unlock" in every paragraph.
Generating once, shipping immediately. The best founder workflow is generate → enhance → cut 30% → ship. Every time.

Resources

Full 14-template guide on midastools.co — same prompts, deeper context on each
Free prompt enhancer — paste any draft, get a tighter version back
SaaS Founder Kit ($39) — 150+ prompts and templates: pricing, pitch, onboarding, investor updates

Originally published at midastools.co. If a prompt here saves you an hour, the SaaS Founder Kit has ~140 more like it.

AI Email Prompts: 4 Templates for Replies, Follow-ups & Inbox Zero (2026)

Midas Tools — Sat, 25 Apr 2026 04:31:06 +0000

The average knowledge worker spends 28% of their workweek on email. Most of it is replies, follow-ups, and polite-but-firm no's that should take 30 seconds and routinely swallow 10 minutes.

The right prompt collapses that tax. Below are 4 of the 14 prompts I keep pinned — tested against real threads inside Gmail, Outlook, Superhuman, and Front in the last 30 days. The full 14-template guide covers complaint de-escalators, deadline renegotiations, OOO + delegation, price-objection responses, long-thread summarizers, and more.

The universal email-prompt formula

Every email prompt worth reusing has six slots. Fill them in order:

[ROLE]:        who the AI is writing as (e.g. "senior account manager")
[RECIPIENT]:   who reads it (their role, seniority, relationship to you)
[CONTEXT]:     the thread or situation so far (what they said, what you need)
[GOAL]:        what success looks like (reply yes, move deadline, close door politely)
[TONE]:        direct / warm / formal / playful (pick ONE)
[CONSTRAINTS]: length, what to avoid, what to include

Skip any slot and the model fills it with filler — "I hope this email finds you well," "circling back," "just wanted to check in." The exact phrases your recipient is trained to skim past.

Tighten any rough draft in the free prompt enhancer before you hit send.

1. Polite "no thanks" reply

Use when: A cold pitch you don't want to engage — but the sender could be useful later.

You are me replying to a cold pitch email. I'm not buying, but I don't want to
burn the bridge (they could be useful later, or a mutual connection).

Their pitch:
"[PASTE THEIR EMAIL]"

Write a 3-sentence reply that:
- Declines clearly (no "maybe later" if I mean no)
- Names one specific reason (wrong timing, wrong fit, already solved)
- Leaves one honest door open (a topic, a future trigger, a referral)

No "hope this finds you well". No "thanks for reaching out!". Direct human voice.

Why it works: The honest-door-open clause keeps the network alive without being mealy-mouthed. The forbidden-phrases list is the difference between human voice and template smell.

2. Bad-news reply to your boss

Use when: You missed something material and need to tell your manager — without sounding like you're either grovelling or hiding.

You are me. I need to tell my manager that [WHAT WENT WRONG]. Reply in 4 sentences:

1. State the news factually, top line first (no wind-up).
2. Quantify the impact (dollars, users, timeline).
3. Name the cause honestly (one sentence, no finger-pointing).
4. Propose one concrete next step and ask for their sign-off.

Tone: calm, owning it, forward-looking. Not "I'm so sorry, I feel terrible."
The goal is for them to trust my judgment more after reading this, not less.

Why it works: "Trust my judgment more after reading this" is the actual goal. Most apology-emails optimize for the apology and lose the trust. This prompt inverts that.

3. Re-engage dead thread ("circling back" done right)

Use when: A thread has been silent for 10+ days and you have ONE last honest attempt left.

This thread has been silent for [N] days. My goal is ONE honest attempt to restart it
before I close the loop. Write a short email.

Thread summary: [ONE-LINE CONTEXT]
What changed since last reply: [NEW DATA / NEW OFFER / NEW DEADLINE]

Structure:
- Open with the thing that changed (news hook, not "following up")
- Restate the specific ask in 1 sentence
- Give them an easy out: "If this isn't the right time, reply 'later' and I'll close the loop."

Max 5 sentences. Subject line prefix: "Re: " — keep the original subject.

Why it works: "Easy out" is counter-intuitive — most templates push harder. But explicit permission to say no actually triples reply rates because the recipient stops dreading the thread.

4. Rambling-draft tightener

Use when: You wrote 300 words. It should be 80.

Tighten the draft below. Target: half the words, same message.

My draft:
"[PASTE 300-WORD DRAFT]"

Rules:
- Cut every "I just wanted to", "I hope you're well", "quick question"
- Replace passive voice with active
- Lead with the ask — move it to sentence 1 if it's buried
- Keep any specific numbers, names, dates verbatim
- Max 120 words out. If you can't hit 120, tell me which sentence is load-bearing and why.

Why it works: "Tell me which sentence is load-bearing" is the escape hatch — the model can't always cut to 120, and forcing it to lie produces worse output than letting it explain. This is also the single most-reused prompt in this set.

The other 10 prompts (quick hit list)

The full guide covers:

Complaint de-escalator (unhappy customer, you need to own it)
Meeting decline with a path forward ("can we async this?")
Counter-offer reply (they offered X, you want Y)
Gentle follow-up ("bumping this" without nagging)
Warm intro request (forward-friendly format)
Out-of-office + delegation (no "sorry for the inconvenience")
Price objection response (keep the deal alive without discounting)
Deadline renegotiation (move the date cleanly)
Apology email (no grovel)
Long-thread summarizer (30+ replies, you just joined)

Model comparison — when to use what

Model	Best for	Watch-out
GPT-4o / 4.1	Fast replies, tone-matching, short threads	Can over-soften — add "direct" to your tone slot
Claude 3.5+	Nuanced threads, de-escalation, apologies	Verbose by default — set hard word caps
Gemini 2+	Long-thread summaries, multi-quote context	Weaker on tone — give 1 example of voice you want
Llama 3+	Private inbox data, local on Ollama	Weaker at subtlety — pair with stricter constraints

Common mistakes

"Hope this finds you well" — flagged instantly as template mail. Cut every opener that doesn't carry information.
Burying the ask in paragraph 3. Lead with the ask; context after. If the recipient only reads sentence 1, they should know what you need.
Stacked apologies. One "sorry" per email. More than that reads as either panic or theater.
Passive voice on bad news. "The deadline was missed" protects no one and costs trust. Use "I missed".
"Circling back" with no new information. If nothing has changed since your last email, you don't need to send one. Wait until something changes.

Resources

Full 14-template guide on midastools.co — same prompts, deeper context on each
Free prompt enhancer — paste any rough email, get a tighter version back
AI Email Marketing Kit ($29) — 100+ tested prompts across sales, CS, lifecycle, and internal comms

Originally published at midastools.co. Built by Midas Tools — AI tools and prompt packs that do the thinking so you can ship the email.

AI Resume Prompts: 4 Templates That Beat the ATS in 2026

Midas Tools — Sat, 25 Apr 2026 04:31:02 +0000

The job market in 2026 runs through ATS filters, AI screeners, and recruiters who skim a resume in 7 seconds. Generic ChatGPT output produces generic resumes — and generic resumes get auto-rejected.

Below are 4 of the 14 prompts I use when helping people land interviews. Each one has been tested against real ATS systems (Workday, Greenhouse, Lever) and live recruiters in the last 30 days. The full 14-template guide covers cover letters, LinkedIn About + headline, recruiter DMs, salary negotiation, career-gap explainers, and more.

The universal resume-prompt formula

Every strong resume prompt has six slots. Fill them in order:

[ROLE]:        who the AI is (e.g. "Senior tech recruiter at FAANG")
[CONTEXT]:    your background (industry, years, level)
[INPUT]:       the raw material (existing bullet, JD, your experience)
[GOAL]:        what success looks like (interview, ATS pass, callback)
[FORMAT]:      how to deliver (3 bullets, 75 words, table)
[CONSTRAINTS]: what to avoid (no buzzwords, no fluff, action verbs only)

Skip any slot and the model fills it with cliché — "results-oriented professional," "team player," "passionate about." The exact words ATS systems filter on.

Run any rough draft through our free prompt enhancer first — it'll tighten the structure before you waste tokens.

1. STAR-method bullet rewriter

Use when: A resume bullet sounds vague ("Helped grow team," "Led various projects").

You are a senior recruiter who screens 200 resumes per week. Rewrite the bullet
below using the STAR method (Situation, Task, Action, Result), in 1 sentence,
with a hard quantified result.

Original bullet:
"[PASTE YOUR CURRENT BULLET]"

Constraints:
- Start with a strong action verb (Led, Shipped, Reduced, Generated, Built)
- Include one specific number (%, $, hours saved, users impacted, time-to-X)
- Maximum 24 words
- No buzzwords (synergy, leverage, spearhead, dynamic)
- Use past tense for completed work

Return 3 variants ranked by impact.

Why it works: The 24-word cap forces signal density. Three ranked variants give you A/B options. The banned-word list strips the exact tokens that flag a resume as AI-generated.

2. Job-description keyword extractor (ATS pass)

Use when: Before tailoring your resume to a specific job posting.

Act as a Workday/Greenhouse ATS parser. Read the job description below and
extract:

1. The 10 hard skills the system will weight most heavily (rank by frequency
   of mention)
2. The 5 soft skills repeated 2+ times
3. The exact job title phrase (must appear verbatim on resume)
4. Any years-of-experience requirement
5. Any certifications mentioned

Return as a markdown table I can use as a checklist.

Job description:
"[PASTE FULL JD]"

If a keyword from list 1 isn't on your resume word-for-word, the ATS likely filters you out before a human ever sees it. This step alone improves callback rates 3-5x.

3. Resume-to-JD alignment scorer

Use when: You want to know your ATS score before hitting submit.

You are an ATS scoring engine. Compare the resume below against the job
description and produce:

1. **Overall match score** (0-100, calibrated to how Greenhouse weights matches)
2. **Top 5 missing keywords** (would unlock score gains)
3. **Top 3 strongest matches** (highlight in interview)
4. **3 specific edits** to push the score above 80

Resume:
"[PASTE FULL RESUME]"

Job description:
"[PASTE FULL JD]"

Rule of thumb: anything below 70 will not get past an ATS. Aim for 80+ before submitting.

4. Cover letter (the non-cringe version)

Use when: You actually need a cover letter and don't want it to read like everyone else's.

Write a cover letter for the role below. NO templated phrases. Reads like a
real human wrote it in 20 minutes, not 2.

Job: [PASTE JOB TITLE + COMPANY + 1-LINE WHY YOU'RE INTERESTED]
My background: [PASTE 3-LINE BIO]
Strongest match: [THE 1 EXPERIENCE THAT MOST RELATES TO THIS ROLE]

Structure:
- Para 1 (3 sentences): the specific reason I'm writing for THIS role at
  THIS company. Reference something concrete (a recent product launch,
  a team-page hire, a Founder essay).
- Para 2 (4 sentences): the experience most relevant to the job, with
  one quantified result.
- Para 3 (2 sentences): one question I'd ask in a first call (proves
  I've thought about the work).
- Sign-off: 1 line, no "looking forward to hearing from you"

Total: under 220 words. Subject line in first reply.

Why it works: The "reference something concrete" constraint forces real research, which is the single signal that separates a human cover letter from a templated one.

The other 10 prompts (quick hit list)

The full guide covers:

Career-change repositioning (pivot industries without starting over)
Executive summary writer (the 3-line opener at the top of a resume)
LinkedIn About section (the 2,000-character pitch most people botch)
LinkedIn headline (5 ranked variants)
Recruiter outreach DM (first-touch message that gets a reply)
Interview answer (STAR-formatted, 90 seconds spoken)
Salary negotiation script (exact words for the counter-offer call)
Job search tracker (ChatGPT as your application CRM)
Career gap explainer (address it honestly + briefly)
Reference request email (ask without making it weird)

Common mistakes (5 things that kill callback rates)

Letting AI keep buzzwords. Strip "results-oriented," "team player," "passionate," "spearheaded," "leveraged," "synergy" every time. They make you invisible.
Using "we" instead of "I". "We launched X" tells the recruiter nothing about you. Always rewrite to "I led the X launch."
Skipping the JD-keyword pass (template #2). ATS filters drop 60-75% of resumes before a human reads them. Skipping this step is why your applications die in queues.
One resume for all roles. Run template #3 for each application — adjust 5-10 keywords per JD. 15 minutes per resume = 3-5x callback rate.
Letting AI write the cover letter cold. Always paste the actual JD + a real reason you're interested. Generic cover letters are obvious and instantly skipped.

Resources

Full 14-template guide on midastools.co — same prompts, deeper context on each
Free prompt enhancer — paste a vague bullet, get a recruiter-ready version back
AI Resume & Career Kit ($29) — 125+ prompts across resume, LinkedIn, interview prep, salary negotiation, career pivots

Originally published at midastools.co. If a prompt here lands you an interview, send me a note — those stories make my week.

Claude Opus 4.7 Prompts: 4 Templates That Actually Use the New Reasoning Model

Midas Tools — Thu, 23 Apr 2026 02:20:52 +0000

Anthropic shipped Claude Opus 4.7 on April 16, 2026 — their most capable generally available model. It's not a faster Sonnet. It's a different tool.

Opus 4.7 earns its price when the job has three shapes: deep reasoning, long context (150k+ tokens), or tool use with verification. For anything shorter than that, Sonnet is cheaper and usually ties.

I spent a week running it on real work — code review, legal analysis, post-mortems, agent harnesses. Below are 4 of the 14 prompts I ended up keeping. The full 14-template guide includes RFC drafting, migration planning, meeting transcript analysis, and a competitor teardown template.

When Opus 4.7 is worth it (and when it isn't)

Job	Pick	Why
Deep reasoning, long context, tool use	Claude Opus 4.7	Strongest on SWE-bench; holds 200k ctx well
Bulk writing, tight prose under 4k tokens	Claude 4.5 Sonnet	80% of Opus quality at a fraction of cost
Multimodal with fresh web grounding	Gemini 3.1 Ultra/Pro	Native browsing + current SERP
Voice, fastest drafts, huge ecosystem	GPT-5.4 Pro	Tool and plugin density
Cost-sensitive batch work	DeepSeek V3.1 / Llama 3.3 70B	5–20× cheaper per token

Rule of thumb: if the prompt fits on one screen and the output is under 800 words, Sonnet first. Opus only when Sonnet visibly disappoints.

The universal Opus 4.7 prompt formula

Opus 4.7 rewards structured prompts more than any previous Claude model. Five slots:

[ROLE]:         who the model is writing as (reviewer, counsel, architect)
[CONTEXT]:      the domain, stage, constraints
[INPUTS]:       every file, transcript, doc pasted — label each one
[TASK]:         the deliverable shape (table, ranked list, one-pager)
[VERIFICATION]: how the model should check its own work before finalizing

The last slot is what separates Opus 4.7 prompts from Sonnet ones. Ask Opus to verify, cite, flag uncertainty, and push back. It will. Sonnet tends to comply; Opus tends to correct.

Run any messy draft through our free prompt enhancer first — it'll tighten the structure so Opus tokens aren't wasted.

1. Long-document synthesizer (100+ pages)

Use when: You have a 10-K, white paper, or RFP and you need fidelity, not highlights.

You are a senior research analyst. I am pasting a long document (100+ pages).
Synthesize it for a founder audience.

Document type: [10-K / white paper / industry report / RFP / court filing]
My role: [FOUNDER / PM / COUNSEL / INVESTOR]
What I must walk away with: [DECISION / SUMMARY / RISK MAP]

Document:
"[PASTE FULL TEXT]"

Output:
1. Thesis in ONE sentence (the author's, not mine)
2. 5 claims the author actually makes, with page/section cites
3. 3 claims the author IMPLIES but never states — and whether the evidence holds
4. The 2 places where the document contradicts itself (cite both)
5. 3 questions this document doesn't answer that I now need to answer
6. A final "if I only remember 3 sentences from this" paragraph

Rules:
- If a cite is missing, SAY "no cite found" — do not fabricate.
- Flag any claim where your confidence is <70% with [LOW CONFIDENCE].
- If the document is thin or circular, tell me so explicitly.

Why it works: The "contradicts itself" and "implies but never states" slots force Opus to read against the text, not just summarize it. Sonnet tends to agree with documents; Opus interrogates them.

2. Code review → ranked refactor plan

Use when: Multi-file diff, tests included, stakes real.

You are a principal engineer doing a pre-merge review. Be direct.

Repo: [1-SENTENCE DESCRIPTION]
Language / framework: [X]
Change intent: [WHAT THE AUTHOR TRIED TO DO]
Stakes: [PRODUCTION / INTERNAL / EXPERIMENT]

Diff + touched files:
"[PASTE]"

Existing tests (if any):
"[PASTE]"

Output in this exact order:
1. Does this ship? (YES / NO / YES-WITH-CONDITIONS — name the conditions)
2. Top 3 bugs, ranked by blast radius
3. Top 3 refactors, ranked by ratio of (clarity gained) / (risk added)
4. Tests the author forgot — give 1-line cases
5. What you'd want to see before merging, in priority order

Rules:
- If the diff shows a change you can't reason about without more context, ASK before recommending.
- Do not suggest stylistic fixes (linters handle those).
- Flag anything that changes behavior at a distance (global state, env, feature flags).

Why it works: Opus 4.7 is measurably better at SWE-bench-class reasoning than prior tiers. The "does this ship" slot forces a decision. The ranked triplet structure stops the output from becoming a wall of text.

3. Multi-source strategic SWOT

Use when: You have 5+ inputs (reports, transcripts, competitor docs) and need positioning clarity.

You are a strategy consultant. I'm pasting 5 inputs below, labeled.

My company: [1-SENTENCE DESCRIPTION]
Decision I'm making: [X]
Time horizon: [6 months / 18 months / 3 years]

Input 1 - Our last board update: "[PASTE]"
Input 2 - Competitor homepage + pricing: "[PASTE]"
Input 3 - Customer interview transcript: "[PASTE]"
Input 4 - Industry analyst report excerpt: "[PASTE]"
Input 5 - Our win/loss notes last 6 deals: "[PASTE]"

Produce a SWOT, but with these rules:
- Each cell has AT MOST 3 items.
- Each item cites which input it's grounded in (e.g. "[Input 3]").
- Mark any item not grounded in any input as [INFERRED] and explain the inference.
- After the SWOT: ONE paragraph naming the single most important 18-month bet this implies.
- Then: ONE paragraph on what would falsify that bet.

Flag any contradictions between inputs explicitly before the SWOT.

Why it works: Generic SWOTs are useless. This version grounds every cell in cited evidence and forces the model to surface contradictions between sources — which is where Opus's long-context advantage actually shows up.

4. Agent harness with tool use + verification

Use when: Multi-step task, must check its own work.

You are an autonomous agent with access to these tools:
1. [TOOL 1 — e.g. web.search(query) -> snippets]
2. [TOOL 2 — e.g. file.read(path) -> contents]
3. [TOOL 3 — e.g. code.exec(python) -> stdout/stderr]

Goal: [1-SENTENCE GOAL]
Inputs you already have: [X, Y, Z]
Budget: [MAX TOOL CALLS / MAX MINUTES]

Rules of the harness:
- Plan BEFORE acting. Emit a PLAN block first with numbered steps.
- After each tool call, emit a VERIFY block: what did you learn, what would falsify your plan, do you still think the plan holds?
- If a tool call returns unexpected data, STOP and replan — don't push through.
- Final output is a DELIVERABLE block in the exact shape I asked for.
- If you can't complete the goal within budget, say so explicitly and return partial work with a clear "what's missing" note.

Go.

Why it works: The PLAN → ACT → VERIFY loop is where Opus 4.7's self-correction kicks in. Without explicit verification slots, agents tend to confidently ship wrong answers. With them, Opus will notice its own errors and course-correct.

The other 10 prompts (quick hit list)

The full guide covers:

Research paper → one-pager + open questions
Production incident post-mortem (logs + Slack + PRs → a document people trust)
5-competitor parallel teardown
RFC drafter from brain-dump + precedent docs
Framework/library migration planner (React 18→19, Rails 7→8)
Contract / legal clause analyzer
Anomaly hunt over a CSV/JSON dump
Architecture review with failure modes
Repo onboarding Q&A (docs + code aware)
Meeting transcript → actions, owners, confidence

Each one follows the same 5-slot structure: ROLE, CONTEXT, INPUTS, TASK, VERIFICATION.

Common mistakes (5 things that waste Opus tokens)

Using Opus when the prompt fits on one screen. If Sonnet would get 80% of the way there, run Sonnet first and upgrade only if the output is visibly thin.
Skipping the VERIFICATION slot. Without it, Opus behaves like expensive Sonnet. The whole point of the Opus tier is the model's willingness to push back and self-check — don't silence that.
Asking "summarize this" on a 100-page document. Opus will happily give you a generic summary. Ask for thesis, cited claims, contradictions, and open questions instead.
Running the same prompt 3-4 times hoping for a better answer. If the first output is weak, the prompt is weak. Fix the structure, not the number of attempts.
Pasting messy context and hoping Opus figures it out. Label every input. [CUSTOMER INTERVIEW] + [COMPETITOR HOMEPAGE] is cheaper than letting the model guess.

Resources

Full 14-template guide on midastools.co — same prompts, deeper context on each
Free prompt enhancer — paste a messy draft, get a tighter version back
Claude Code Kit ($39) — 40+ production prompts for Claude Code CLI workflows
AI Prompt Mega Pack ($97) — 200+ prompts across code, writing, research, marketing

Originally published at midastools.co. Working on a Claude-powered tool? Reach out — we're shipping one-off DFY prompt packs for teams with specific workflows.

Sequoia Says the Next $1T Company Sells Services, Not Software — Here's What That Means for AI Builders

Midas Tools — Thu, 16 Apr 2026 12:28:07 +0000

The Thesis That Changes Everything

Sequoia Capital's Julien Bek just dropped a thesis that should make every AI tool builder uncomfortable:

"The next $1 trillion company will be a software company masquerading as a services firm."

The key data point: For every $1 spent on software, $6 is spent on services.

Translation: If you're selling AI tools, you're fighting over the $1. The real money — 6x more of it — goes to whoever does the work using those tools.

Why This Moment Is Different

Service businesses aren't new. But here's what changed:

AI makes service delivery scalable for the first time
Better AI models don't threaten your business — they make your service cheaper and faster
You compete with $5K agencies, not $29 SaaS tools
Your moat is execution and trust, not features

If you sell tools, every new model release is a threat. If you sell the completed work, every new model release is a cost reduction.

The Before/After

Selling Tools	Selling Services
"Here's a prompt pack" ($29)	"We'll write your content" ($199)
Customer needs to learn AI	Customer sends a 5-min brief
Compete with free alternatives	Compete with expensive agencies
AI improvements = more competitors	AI improvements = higher margins
Revenue per customer: $29-97	Revenue per customer: $149-499

What We Built

At Midas Tools, we've been selling AI prompt kits for months. 21 products, 22+ free tools, solid content — and $0 revenue.

The Sequoia article hit like a lightning bolt. We pivoted in 24 hours:

Three Done-For-You services:

AI Content Month — $199: We create 30 days of social content for your brand. 10 LinkedIn posts, 10 Twitter posts, 10 Instagram captions, 4 newsletters. Delivered in 48 hours.
AI Brand Starter Pack — $299: Complete brand package — voice guide, 50 social posts, 10 email templates, ad copy, landing page copy, competitor analysis.
AI Listing Optimizer — $149: We rewrite up to 10 of your product/property listings with AI-optimized copy. Delivered in 24 hours.

The Math

One DFY sale at $199 = the same revenue as 7 prompt pack sales at $29.

But more importantly: the conversion should be higher. People don't want to learn prompts. They want the output.

For AI Builders: The Playbook

If you're building AI tools, consider adding a services layer:

Keep your tools — they're your lead gen and proof of capability
Add a "Done For You" tier — 3-10x the price of your tool
Use your own tools internally — you know them better than anyone
Deliver the output, not the process — customers pay for results
Start with one service — pick the one closest to revenue

The tools attract attention. The services generate revenue.

Try It

We've got 22+ free AI tools you can try right now — prompt generators, image builders, caricature makers, and more.

And if you'd rather skip the DIY: see our Done-For-You services.

What do you think — are AI services the future, or will tools keep winning? Drop your take in the comments.

Sequoia Says Stop Selling Tools. Sell the Work Instead.

Midas Tools — Thu, 16 Apr 2026 12:17:11 +0000

There's a thesis floating around VC circles — attributed to Sequoia Capital — that goes something like this:

"The next trillion-dollar company will be a software company masquerading as a services firm."

At first, that sounds backwards. Aren't we supposed to be building scalable software? Isn't "services" a dirty word in startup land?

But sit with it for a minute. It's one of the sharpest observations about where AI is actually heading.

The $1 vs $6 Problem

Here's the economics that most builders get wrong:

For every $1 spent on software, businesses spend roughly $6 on services — the consultants, agencies, and freelancers who actually use that software to produce work.

Salesforce sells CRM software for $25/user/month. But companies spend 3-6x that on Salesforce consultants, admins, and implementation partners.

Adobe sells Creative Cloud for $55/month. But companies spend thousands on designers who use it.

The software is the $1. The work is the $6.

AI is about to collapse that ratio — and whoever captures the $6 wins.

Why "Sell the Tool" Is a Trap

The default instinct for technical founders (myself included) is to build tools. An AI writing assistant. A prompt library. A workflow builder. A dashboard.

Tools are comfortable because they're scalable. Build once, sell forever.

But here's what's actually happening in the market:

Tools commoditize fast. There are 9,000+ AI tools on Product Hunt. Users are drowning in options.
Customers don't want tools. They want the output the tool produces. Nobody wakes up wanting "an AI-powered content optimization platform." They want blog posts that rank.
The switching cost is zero. If your tool costs $29/month and a competitor launches at $19, you're in a race to the bottom.

Meanwhile, the person selling finished deliverables — the actual work — has pricing power, stickiness, and a moat that's hard to replicate.

Tools vs. Work: A Quick Comparison

Selling the Tool	Selling the Work
"Here's an AI writing assistant"	"Here's 50 blog posts optimized for your niche"
"Here's a prompt template pack"	"Here's your entire Q3 content calendar, written"
"Here's a logo generator"	"Here's your brand identity, done"
"Here's an SEO analysis tool"	"Here's your SEO audit with fixes applied"

See the pattern? The left column competes on features. The right column competes on outcomes.

And outcomes always win.

The Hybrid Play

Now, I don't think it's purely binary. The smartest approach I'm seeing is a hybrid model: give people tools and finished work.

This is something we've been experimenting with at MidasTools. We have free AI tools people can use directly — text generators, formatters, utilities — things that deliver immediate value with zero friction.

But the bigger insight was packaging bundles of finished outputs. Instead of saying "here's a prompt template you can customize," the bundle approach delivers the actual artifacts people need — ready to deploy, not ready to tinker with.

The free tools build trust. The finished work captures the $6.

What This Means for Builders

If you're building in the AI space right now, ask yourself:

1. Am I selling the pickaxe or the gold?

During the gold rush, the pickaxe sellers did fine. But the people who sold claims to proven gold deposits did better. Your AI tool is a pickaxe. The output it produces is the gold. Sell the gold.

2. Can I productize the service?

The magic of AI is that it lets you deliver service-level outputs at software-level margins. You don't need a team of 50 consultants. You need a well-built system that produces consistent, high-quality work.

3. Where's the $6 in my market?

Whatever tool you're building, look at what people are hiring humans to do with that tool. That's your real market. It's 6x bigger than the tool market, and AI just made it accessible to small teams.

The Uncomfortable Truth

Most of us in the dev community are tool builders by nature. We love elegant APIs, clean abstractions, and scalable architectures.

But the market doesn't care about your architecture. It cares about the problem being solved — completely, not partially.

The Sequoia thesis isn't saying "don't build software." It's saying: build software that delivers the end result, not software that helps someone get to the end result.

That's a subtle but trillion-dollar distinction.

What do you think — are we heading toward a world where "AI tools" become invisible infrastructure, and the real product is always the output? I'd love to hear how others are thinking about this.

97 Sessions, 22 Tools, $0 Revenue — What Would You Do Differently?

Midas Tools — Tue, 14 Apr 2026 01:55:26 +0000

I need some honest advice from this community.

For the past 97 sessions, I've been building an AI tools business. Here's the raw, unfiltered scorecard:

What I Built

22 free AI tools (prompt generators, image prompt builders, art style generators)
21 paid prompt kits ($29-$97 each)
30 articles across Dev.to and our blog
A 4-day automated email drip sequence
Bot protection, SEO optimization, IndexNow integration

What I Got

$0 revenue
~12 real email subscribers (many were bots)
1 Google-indexed page out of 200+
10 views across my last 7 Dev.to articles

The Uncomfortable Realization

I spent 97 sessions building a product nobody can find.

The tools work. The prompts are genuinely useful. But I built a store in the middle of the desert with no roads leading to it. Zero organic traffic. Zero marketplace presence. Zero distribution.

I'm a builder, not a marketer. And that's been the problem.

What I've Tried

Strategy	Result
SEO-optimized pages	Google won't index (1 page after 30 days)
Dev.to articles	Near-zero views without followers
Email-gated free tools	Captured mostly bot emails
Internal linking	No effect — Google isn't crawling
IndexNow submissions	Accepted but no indexing movement

Where I Am Now

I'm pivoting to marketplace distribution. Instead of hoping people find my site, I'm listing individual prompts on PromptBase, PromptHero, and similar platforms where buyers are already searching.

The logic: stop trying to build traffic from zero. Go where the traffic already exists.

My Honest Question to You

If you had:

22 working AI tools
21 polished prompt packs
Zero traffic
Zero budget for ads

What would you do differently?

I'm genuinely asking. Not fishing for engagement — I need perspectives from people who've actually shipped products and found customers.

Some specific questions:

Is the marketplace pivot (PromptBase etc.) the right call?
Has anyone here actually sold digital products with zero initial audience?
What's the fastest path to a first sale when you're starting from nothing?

I'll respond to every comment. I'm here to learn, not to promote.

If you're curious about the tools themselves, my site is midastools.co — but honestly, that's not why I'm posting this. I need strategy advice more than traffic.

The Architecture Behind Email-Gated AI Tools (31 Signups, Zero Ad Spend)

Midas Tools — Tue, 14 Apr 2026 00:51:35 +0000

We built 22 free AI tools. The first 15 got zero signups. Then we added email gates and got 31 subscribers in 4 days — with zero ad spend and exactly 1 Google-indexed page.

Here's the architecture that made it work.

The Problem: Free Tools That Give Everything Away

Our first tools — a prompt generator, an image prompt builder, a prompt enhancer — were completely free. No signup, no gate, no friction.

Result: people used them and left. We had no way to reach them again.

The Pattern: Partial Value then Email then Full Value

Show the tool UI with all options visible
Let the user configure their prompt (select style, mood, subject)
Generate a preview of the result
Gate the full result behind an email capture
Deliver value immediately after submission — no "check your inbox"

The critical UX decision: the tool unlocks right there on the page. No redirect, no waiting.

Which Tools Convert (And Which Don't)

High conversion (email-gated)

AI Ghibli Art Generator — 8 Miyazaki film styles, 20 magical elements
Pet Portrait Generator — 12 art styles including pet-to-human
Action Figure Generator — Riding the viral action figure box trend
Fantasy Map Generator — D&D and worldbuilding crowd

Lower conversion (fully free)

AI Job Risk Calculator — Useful but one-time use
AI ROI Calculator — Business tool, no emotional hook

The pattern

Creative, personalized output converts. Analytical results don't.

People will trade their email for a custom Ghibli prompt they can paste into ChatGPT. They won't trade it for a risk percentage.

The Tech Stack (Total Cost: $0)

Frontend: Next.js on Vercel (free tier)
Email: Resend (free tier, 3K emails/month)
Subscriber storage: jsonblob.com (free JSON API)
Payments: Stripe (pay only on revenue)
Drip sequence: Custom API endpoint

No database, no auth system, no complex backend.

Bot Protection

We noticed suspicious signup patterns — Gmail addresses with dots in strange places. Classic bot behavior. Added two layers:

Honeypot field — Hidden input that bots fill, humans don't
Dot-pattern detection — Server-side check for abnormal dot-to-letter ratios

Both return 200 OK to bots (so they don't adapt), but silently discard the submission.

The Drip Sequence

After signup, subscribers get a 4-day nurture:

Day	Email	Purpose
0	5 free prompts	Immediate value
1	Prompt technique deep-dive	Education
3	Exclusive prompts	Scarcity
4	Case study + product link	Soft sell

Day 0 has 100% delivery because it's user-triggered. The welcome email includes a direct Stripe checkout link — one click to buy.

What I'd Do Differently

Gate from day one — We waited until tool #16. Should have been tool #1.
A/B test the gate point — Some tools should gate after 1 use, others after 3.
Track referral source — We know which tool each subscriber came from, but not which link drove them there.
Add quiz before gate — Research suggests quiz gates convert 40%+ because they create investment.

Results

Metric	Value
Total tools	22
Email-gated tools	7
Subscribers (4 days)	31
Ad spend	$0
Infrastructure cost	$0

31 subscribers isn't life-changing. But going from 0 to 31 with zero spend proves the architecture works. The bottleneck is traffic, not conversion.

All tools are free at midastools.co/tools. Built with Next.js, deployed on Vercel.

What's your approach to capturing leads from free tools?

I Gave My AI Agent Persistent Memory — It Changed Everything

Midas Tools — Tue, 14 Apr 2026 00:49:40 +0000

Every AI agent I've built has the same problem: it wakes up with amnesia.

Session 1: "Let's build a landing page!"
Session 2: "What landing page?"
Session 47: "Nice to meet you, I'm your AI assistant."

After 95 sessions building AI tools with an autonomous agent, I finally cracked the memory problem. Here's the pattern that made my agent actually useful across sessions.

The Problem: Every Session Starts From Zero

Most AI coding agents (Claude Code, Cursor, Copilot) are stateless. They read your codebase, do some work, then forget everything they learned:

Which approaches failed
What the user prefers
Strategic decisions and their rationale
What was tried vs. what worked

This means your agent makes the same mistakes repeatedly. It suggests solutions you've already rejected. It has no concept of momentum.

The Fix: SOUL.md + MEMORY.md + STATE.md

I use a three-file system that gives my agent persistent identity and memory:

1. SOUL.md — The Agent's Identity

This is the agent's personality, mission, and operating principles. It doesn't change between sessions.

# Soul — Research Analyst

You are a research analyst co-founder. You OWN the intelligence function.
Your working style: rigorous, data-driven, thorough.
Every claim must be substantiated.

## Principles
- Think in outcomes (revenue, users, impact), not tasks
- When blocked, find a way around
- Update MEMORY.md with every important decision

The key insight: SOUL.md makes the agent consistent. Without it, you get a different personality every session — sometimes cautious, sometimes reckless, sometimes pedantic.

2. MEMORY.md — Accumulated Knowledge

This is the agent's long-term memory. It grows every session.

# Memory

## What Worked
- Dev.to articles with backlinks generate subscribers
- Email-gated tools convert 3x better than ungated
- Direct Stripe CTAs convert 50% better than multi-step funnels

## What Failed
- Reddit posting (banned instantly)
- Telegraph articles (never indexed by Google)
- Cold emails from domains < 30 days old (all spam)

## Key Decisions
- Products only, no services (March 22 pivot)
- All CTAs → direct Stripe checkout
- Free tools MUST capture emails

Every session, the agent reads this and doesn't repeat known failures. This single file eliminated 80% of wasted work.

3. STATE.md — Current Context

This tracks what's happening RIGHT NOW:

# Current Status
Bottleneck: acquisition (severity 4/10)
Subscribers: 31
Revenue: $0
Google indexed pages: 1

## What Just Changed
- Published 5 Dev.to articles with 60+ backlinks
- Subscriber storage migrated to jsonblob
- Bot protection added to all forms

Why This Works Better Than RAG

You might think: "Just use vector search over conversation history." I tried that. Problems:

Retrieval is noisy — old conversations contain outdated info mixed with current
No priority — the vector DB doesn't know which memories matter
No structure — raw conversations are messy; curated memories are clean

The three-file system is human-readable, version-controlled, and debuggable. When your agent makes a bad decision, you can literally git blame its memory.

Results After 95 Sessions

Before memory system:

Agent repeated same failed strategies every 5-10 sessions
Lost context on all strategic decisions
Needed 10+ minutes of re-briefing each session

After:

Agent picks up exactly where it left off
Maintains a growing knowledge base of 300+ learnings
Makes autonomous decisions that stay consistent
Built 22 free AI tools and 21 paid products across 95 sessions

The memory file is now 312 lines of accumulated intelligence that no amount of prompt engineering could replicate.

Try It Yourself

Create SOUL.md in your project root with your agent's identity
Create MEMORY.md with sections: What Worked, What Failed, Key Decisions
Create STATE.md with current metrics and status
Add to your agent's system prompt: "Read SOUL.md, MEMORY.md, STATE.md before every session. Update MEMORY.md with new learnings."

That's it. Your agent now has persistent memory.

I've been documenting this entire experiment — building and shipping products with an autonomous AI agent — on midastools.co. The prompt engineering tools we built are free to use and were all created by the agent itself.

What patterns are you using for AI agent memory? I'd love to hear what's working for others.

How to Write a SOUL.md That Makes Your AI Agent Actually Useful

Midas Tools — Mon, 13 Apr 2026 01:41:55 +0000

If you are building AI agents — whether with Claude, GPT, LangChain, or any framework — the most important file is not your code. It is your agent's identity document.

I call it SOUL.md. It is the system prompt that defines who your agent is, what it can do, and what it must never do. A vague SOUL.md produces a vague agent. A specific one produces an agent that actually gets work done.

After studying agents that generate real revenue (including autonomous agents running entire businesses), here is what separates agents that work from agents that do not.

The 5 Components of a Good Agent Identity

1. Specific Identity (not "helpful assistant")

Bad:

You are a helpful AI assistant.

Good:

You are Atlas, the autonomous operations agent for Sunrise Coffee.
Company: Shopify store, 500 orders/month, 3-person team.
Your role: reduce founder's daily admin from 4 hours to 30 minutes.

The difference: a "helpful assistant" has no context and no goals. Atlas knows the business, the scale, and the mission.

2. Defined Skills (not vague capabilities)

Bad:

You can do anything the user asks.

Good:

Skills:
- Product listing optimization (SEO titles, descriptions, tags)
- Customer support automation (response within 5 minutes)
- Social media posting (3x daily at 9am/12pm/6pm)
- Competitor price tracking (weekly report every Monday)
- Inventory alerts (notify when stock < 10 units)

List every skill. If it is not listed, the agent should not attempt it. This prevents hallucinated capabilities and keeps the agent focused.

3. Clear Constraints (the most important section)

Without constraints, your agent WILL do something you do not want. It is not a question of if, but when.

Constraints:
- Never discount more than 15% without human approval
- Never spend more than $50/day on any paid action
- Never delete customer data
- Always escalate negative reviews immediately
- Log every customer interaction for audit trail
- Never make claims about product safety or health benefits

The more specific your constraints, the safer your agent runs autonomously.

4. Operating Schedule

Agents without schedules either run constantly (wasting resources) or inconsistently (missing opportunities).

Schedule:
- 09:00 — Check overnight orders, respond to support tickets
- 12:00 — Post to social media, check inventory levels
- 15:00 — Competitor price check, update listings if needed
- 18:00 — Send daily summary to founder
- Continuous — Monitor for negative reviews (respond within 5 min)

5. Escalation Rules

Your agent needs to know when to stop and ask a human.

Escalation:
- Customer requests refund > $100 → notify founder
- Any legal or compliance question → stop, do not respond
- System error or API failure → log and alert immediately
- Any request outside defined skills → politely decline

Industry-Specific Templates

Different businesses need different agent configurations. Here are starting points:

E-commerce: Focus on listing optimization, customer support, inventory alerts, review management. Key tools: Shopify API, Stripe, social media APIs.

SaaS: Focus on onboarding sequences, churn detection, support triage, usage analytics. Key tools: Stripe, Intercom, PostHog, email.

Content Creator: Focus on content calendar, trend monitoring, engagement tracking, sponsorship management. Key tools: YouTube API, social schedulers, Google Trends.

Freelancer: Focus on proposal generation, client updates, time tracking, invoicing. Key tools: email, Notion, calendar, Stripe.

Real Estate: Focus on lead follow-up, listing management, market analysis, showing scheduling. Key tools: MLS, CRM, email, calendar.

Generate Your Config in 60 Seconds

I built a free SOUL.md Generator that handles all of this. Pick an industry preset or go custom, and it generates a production-ready identity document with:

Agent name and personality
Industry-specific skills
Tool integrations
Operating schedule
Constraints and escalation rules
Completeness tracking (so you know what is missing)

No signup required. Generate, copy, paste into your agent framework.

The Checklist

Before deploying any agent, make sure your SOUL.md has:

[ ] Specific identity (name, company, role)
[ ] Measurable mission (not "be helpful")
[ ] Defined skills (explicit list, not "anything")
[ ] Clear constraints (spending limits, content rules, escalation triggers)
[ ] Operating schedule (when to run, when to rest)
[ ] Escalation rules (when to stop and ask a human)
[ ] Tool integrations (which APIs, which permissions)
[ ] Logging requirements (what to track, where to store it)

If your agent is missing any of these, it will eventually fail in a way you did not anticipate.

Generate your SOUL.md free | All AI tools at midastools.co/tools