astraimpactTH/EN
A FIELD GUIDE TO THE NEXT CHAPTER OF WORKVOL. 01 / 2026

When AI can finish the work,
where do you create value?

An independent look at work after GPT-6 Astra.
What changes. What deserves thought. Where to go next.

Explore your work↗
MADE FOR THAI KNOWLEDGE WORKERSStart with the context ↓
A metal arch, stone planes and an orange sphere: an editorial sculpture of work in transition
FIG. 01
HUMAN × MACHINE
A NEW SPACE OF POSSIBILITY01 — 06

THE SHIFT / CONTEXT

From giving answers.
To doing the work.

Astra supports multistep work across code, browsers and professional tools. Results depend on the model, data, access and verification. ↗ ASTRA

1/4

of global jobs have GenAI exposure

ILO estimates potential exposure, not the number of people who will lose jobs. ↗ ILO

22.9%

of employment in ASEAN

has more than minimal exposure in ILO’s estimates for 2025. This is not a Thailand-only or Astra-specific statistic. ↗ ASEAN

No evidence here establishes that Astra has eliminated an entire occupation. Research, demonstrations and editorial analysis are labelled separately.

01 WORK IN TRANSITION

Where your work
starts to change.

Choose a field and inspect its tasks. This editorial analysis is a starting point for testing your own work.

Examine the tasks.
A job title is not a forecast.

FIELD 01 / 06

From volume to outcomes

Template copy and summaries become easier to produce. Clients may pay less per deliverable. The useful question is how your work improves decisions or sales.

Captions / product descriptionsEasier to automate

Clear product data and a brand guide support batch drafts. Check claims, prices and Thai tone.

Campaign reportingWorkflow changes

Agents can assemble metrics and draft commentary. Attribution and causal conclusions need scrutiny.

Customer interviews / positioningHuman accountability

Access actual buyers, weigh evidence and choose which market bet to make.

WHAT COULD DISAPPEAR IN ITS CURRENT FORM

Per-piece template SEO and caption services could lose their market when clients can produce acceptable work themselves.

Scenario: requires lower total costs, acceptable quality and workable adoption. No established extinction timeline.
See the 90-day plan for this field ↗

FIELD 02 / 06

Taste you can articulate

First passes and preparation can become much faster. Direction, deliberate editing and consistency across a brand remain valuable responsibilities.

Resizing / asset variantsEasier to automate

Suitable for batch workflows with clear originals and specifications. Check overflow and proportions.

First-pass edits / 3D draftsWorkflow changes

Creator demos exist. Continuity, accuracy and editable-file quality need testing on your own briefs.

Art direction / client feedbackHuman accountability

Translate business intent into a visual system and explain composition and pacing decisions.

WHAT COULD DISAPPEAR IN ITS CURRENT FORM

Undirected resizing and template-edit packages could be absorbed into software features.

Scenario: requires lower total costs, acceptable quality and workable adoption. No established extinction timeline.
See the 90-day plan for this field ↗

FIELD 03 / 06

From code to reliable systems

Prototypes can get cheaper. Integration, real data and incidents still require ownership. Being able to verify and operate a system matters.

Specified pages / CRUDEasier to automate

Bounded tasks with acceptance criteria fit agents. Verify actual behaviour.

API integration / refactoring / testsWorkflow changes

Results depend on access, documentation and tests of genuine requirements.

Architecture / incidents / securityHuman accountability

Choose trade-offs, assess impact and own production operation beyond a passing build.

WHAT COULD DISAPPEAR IN ITS CURRENT FORM

Template-page services without integration or maintenance may face stronger price competition.

Scenario: requires lower total costs, acceptable quality and workable adoption. No established extinction timeline.
See the 90-day plan for this field ↗

FIELD 04 / 06

Ask decision-relevant questions

Collection and slide drafting can accelerate. Weak evidence can still yield convincing prose. Analysis must make the knowns and unknowns visible.

Information gathering / slide formattingEasier to automate

Works best with traceable public information. Open the original sources.

Spreadsheet analysis / draft forecastsWorkflow changes

Check assumptions, missing data and formulas before interpreting relationships.

Experiment design / decision adviceHuman accountability

Identify decision-changing evidence, choose a test and communicate uncertainty.

WHAT COULD DISAPPEAR IN ITS CURRENT FORM

Collect-and-format reports based only on public information may become harder to sell as standalone work.

Scenario: requires lower total costs, acceptable quality and workable adoption. No established extinction timeline.
See the 90-day plan for this field ↗

FIELD 05 / 06

Own exceptions and relationships

Repeated answers and data transfer suit automation when information is ready. Exceptions and customers needing real help require accountable owners.

Data entry / scheduling / FAQsEasier to automate

Needs rules, access controls and escalation when information is incomplete.

Ticket triage / follow-upWorkflow changes

Use systems for prioritisation; review cases affecting customers or spending.

Complaints / service designHuman accountability

Understand context, negotiate remedies and change the process to prevent recurrence.

WHAT COULD DISAPPEAR IN ITS CURRENT FORM

Roles limited to copying data or reading scripts may shrink when systems connect and escalation works.

Scenario: requires lower total costs, acceptable quality and workable adoption. No established extinction timeline.
See the 90-day plan for this field ↗

FIELD 06 / 06

Verify accuracy, own consequences

Document reading and comparison may accelerate, while mistakes remain costly. Factual, regulatory and verification expertise still matters.

Invoice extraction / document organisationEasier to automate

Readable inputs help automation. Reconcile outputs with originals and records.

Contract comparison / draft reportsWorkflow changes

Use drafts with source references. Check completeness and context every time.

Interpretation / sign-off / case adviceHuman accountability

Qualified, authorised professionals must verify the applicable context and requirements.

WHAT COULD DISAPPEAR IN ITS CURRENT FORM

Pure filing and extraction roles may contract. Drafting capability alone does not remove verification and sign-off responsibilities.

Scenario: requires lower total costs, acceptable quality and workable adoption. No established extinction timeline.
See the 90-day plan for this field ↗

Exposure research informs the framework; individual task judgments are this publication’s analysis. ↗ ILO ↗ BOT

02 MADE WITH ASTRA

See what people are making.

Projects from other creators. Explore the original videos alongside their limits.

Astra inside Final Cut — Ben DavisCREATIVE WORK
01 / CREATOR DEMOSEP 2026

Astra inside Final Cut

The creator reports footage import, colour work and audio syncing. A useful example of preparation and first-pass editing.

Creator-reported result. Quality and time savings need testing on your own footage.

Ben Davis ↗
Listing photos into a 3D walkthrough — Yunfan Ye3D & PROPERTY
02 / CREATOR DEMOSEP 2026

Listing photos into a 3D walkthrough

A property listing becomes a house model and video, illustrating an early visualisation workflow.

The creator acknowledges incorrect details. It is not a verified architectural plan or measurement.

Yunfan Ye ↗
A game project built with Astra — Riley BrownSOFTWARE & GAMES
03 / CREATOR DEMOSEP 2026

A game project built with Astra

A screen recording of a game project illustrates software creation and iteration in a developer workflow.

One project example, not a benchmark of production reliability.

Riley Brown ↗

Preview images belong to the creators · Captions and media metadata checked via a public-post mirror · Projects not reproduced · Loading an embed connects to X. ↗ DEMOS

03 YOUR NEXT MOVE

90 days.
Start with your next task.

A plan to try alongside your work. Measure what happens, then adjust.

  1. 01—30DAYS

    Find your baseline

    Baseline one recurring job across five deliverables. Compare time and revisions with AI on the same briefs.

  2. 31—60DAYS

    Build a workflow

    Interview three customers. Run one small campaign from the findings and measure the agreed outcome.

  3. 61—90DAYS

    Demonstrate value

    Publish a case study with costs, outcomes and mistakes. Test a clearly scoped project offer.

Portfolio: brief → customer evidence → decision → campaign result

  1. 01—30DAYS

    Find your baseline

    Time preparation work. Test on assets you can use and inspect every output.

  2. 31—60DAYS

    Build a workflow

    Build one visual system with repeatable criteria for imagery, colour and pacing.

  3. 61—90DAYS

    Demonstrate value

    Complete one real project. Document before/after, revisions and decisions that improved it.

Portfolio: direction board + editable files + decision rationale

  1. 01—30DAYS

    Find your baseline

    Assign a small feature in a test project. Write acceptance criteria first and review the diff.

  2. 31—60DAYS

    Build a workflow

    Practise integration with logging, rollback and malformed-data handling.

  3. 61—90DAYS

    Demonstrate value

    Let real users try a small project. Record failures, costs and maintenance effort.

Portfolio: working system + meaningful tests + incident notes

  1. 01—30DAYS

    Find your baseline

    Keep a research log of questions, sources, dates and limits for major claims.

  2. 31—60DAYS

    Build a workflow

    Produce one analysis with sensitivity checks. Ask a peer to challenge it.

  3. 61—90DAYS

    Demonstrate value

    Write a decision memo for a real user, including evidence and what would change the recommendation.

Portfolio: decision memo + sources + inspectable assumptions

  1. 01—30DAYS

    Find your baseline

    Map 20 repeated cases using non-identifying data. Separate routine work from exceptions.

  2. 31—60DAYS

    Build a workflow

    Build a small knowledge base. Test difficult questions and human escalation.

  3. 61—90DAYS

    Demonstrate value

    Pilot a revised process and track resolution time, reopen rates and customer satisfaction.

Portfolio: process map + escalation rules + pilot results

  1. 01—30DAYS

    Find your baseline

    Test extraction on synthetic documents. Record error types and review criteria.

  2. 31—60DAYS

    Build a workflow

    Build a workflow linking conclusions to documents and requiring human approval.

  3. 61—90DAYS

    Demonstrate value

    Have a professional review the pilot. Measure savings after review and correction time.

Portfolio: audit trail + error log + review checklist

Progress stays in this browser. Your selections are not sent to a server.

Count review and correction time.
Faster output must still be useful.

04 THE BIGGER PICTURE

When work changes,
the economy follows.

Who receives productivity gains depends on competition, access and policy choices.

A conceptual Bangkok-inspired architectural model in paper and pale stone
FIG. 02 / A SHARED FUTUREAI-created editorial artwork

More output.
Who gets the income?

Lower digital-service costs can help Thai teams sell abroad. Standardised work also faces price competition. Income may accrue to owners of platforms, data and customer access.

These are mechanisms to track, not an Astra GDP forecast. Outcomes also depend on demand, energy costs, investment and national preparedness. ↗ IMF

SCENARIO / FOR PREPARATION

The gap between firms can widen

Assumption: large firms have better data and budgets than SMEs. Output accelerates, but integration and skills constrain adoption. Gains concentrate in some firms.

WHAT YOU CAN TRY

Develop sector expertise and show results from one working, bounded workflow.

Watch: SME adoption / tool costs / output per hour

SCENARIO / FOR PREPARATION

The first rung can get narrower

Assumption: agents handle drafts and firms reduce junior hiring even with stable sales. New workers get fewer opportunities to build experience. This is not a claim about every Thai occupation.

WHAT YOU CAN TRY

Build for real users, document verification and seek expert feedback.

Watch: junior vacancies / paid internships / time to first job

SCENARIO / FOR PREPARATION

The cost of starting can fall

Assumption: tools are accessible and markets competitive. Small Thai teams can sell specialised services abroad more easily. Rivals have the same tools; customers and distribution determine revenue.

WHAT YOU CAN TRY

Test a narrow offer for one customer segment. Measure profit after tool and review costs.

Watch: new businesses / service exports / freelance prices

US evidence suggests watching junior hiring, but cannot establish Thai effects. ASEAN research has not found widespread disruption. ↗ STANFORD ↗ ASEAN

PUBLIC CHOICES

What organisations
and governments can do.

This publication’s proposals for Thailand. Pilot them and publish results before scaling.

↗ IMF ↗ BOT
01

Keep a route into work

+

Employers can offer paid apprenticeships with real problems and reviewers, measuring skill growth alongside output.

02

Make transitions workable

+

Public and employer support can combine learning time, transition income and matching to real vacancies. Measure subsequent employment and earnings.

03

Give SMEs access to tools and markets

+

Pilot shared tools and advice, measure total costs, reduce single-vendor dependence and help firms reach overseas customers.

04

Make public-sector AI accountable

+

Start with verifiable internal tasks, keep audit trails, assign accountable owners and provide a route for people to request review.

05 NOTES & SOURCES

Read further. Check the evidence.

Checked 6 September 2026. Most studies predate Astra and provide context, not direct evidence of this model’s effects.

RESEARCH

Published findings with scope, population and dates.

DEMONSTRATION

Creator examples do not establish general success rates.

ANALYSIS

Our interpretation and scenarios, to be tested on real work.

  1. [01]
    OpenAI · GPT-6 Astra ↗

    Developer documentation on capabilities and tools. Vendor documentation, not a measurement of employment.

  2. [02]
    ILO · GenAI and jobs ↗

    Global occupational exposure, not an individual probability of job loss. Published before Astra.

  3. [03]
    ILO · ASEAN labour markets ↗

    ASEAN context: uneven adoption and preparedness; no widespread GenAI labour disruption established.

  4. [04]
    Stanford · Canaries in the Coal Mine? ↗

    US observational evidence, not a causal estimate or a Thai employment rate.

  5. [05]
    Bank of Thailand · AI & labour ↗

    Thai context on skills and structural challenges; not evidence of Astra-specific effects.

  6. [06]
    IMF · Sharing the gains of AI ↗

    Policy discussion of social protection, education and taxation to distribute AI gains.

  7. [07]
    explainx · Astra demo source index ↗

    Used to discover original posts. Captions and video metadata checked via a public-post mirror; projects not reproduced.

About this report and its artwork +

Astra Impact is an independent publication for Thai knowledge workers, unaffiliated with OpenAI. Content is AI-assisted research synthesis and editorial analysis. The sculpture and city are AI-created illustrations, not evidence or photographs of actual events. Video previews belong to their credited creators. Career and income outcomes are not guaranteed.

THE NEXT CHAPTER IS YOURS TO WORK ON.

Start with one task.
See how much better you can make it.

Build your next move ↗